SlideShare uma empresa Scribd logo
1 de 49
Baixar para ler offline
Activity Recognition from Accelerometers in
Smart Homes
Tom Diethe
Manager, Core Machine Learning
Amazon Research Cambridge
Honorary Research Fellow
Intelligent Systems Laboratory
University of Bristol
Bristol Robotics Lab, 16th August 2017
Introduction Methods Bayesian Dictionary Learning Infer.NET implementation Experiments Conclusions References
About Me
1996-2000 B.Sc. Experimental Psychology @ UoB
2000-2005 QinetiQ Ltd
2005-2006 M.Sc. Intelligent Systems @ UCL
2006-2010 Ph.D. Machine Learning & Signal Processing
2010-2012 Post-doc x2 @ UCL (CS + Stats)
2012-2013 Consultant @ BMJ Group
2013-2014 Researcher @ Microsoft Research Cambridge
2014-2017 Research Fellow @ UoB (SPHERE)
Next stop: Amazon Research Cambridge
Tom Diethe Amazon Research Cambridge; University of Bristol
Activity Recognition from Accelerometers in Smart Homes
Introduction Methods Bayesian Dictionary Learning Infer.NET implementation Experiments Conclusions References
The SPHERE project
Environmental
temp, light, humidity, air quality, water & electricity
Video
emotion, gait, activity, interaction
Wearable
activity, sleep, etc.
Contextual information
demographics, medical history
diaries, annotations
Tom Diethe Amazon Research Cambridge; University of Bristol
Activity Recognition from Accelerometers in Smart Homes
Introduction Methods Bayesian Dictionary Learning Infer.NET implementation Experiments Conclusions References
The SPHERE project
Environmental
temp, light, humidity, air quality, water & electricity
Video
emotion, gait, activity, interaction
Wearable
activity, sleep, etc.
Contextual information
demographics, medical history
diaries, annotations
Tom Diethe Amazon Research Cambridge; University of Bristol
Activity Recognition from Accelerometers in Smart Homes
Introduction Methods Bayesian Dictionary Learning Infer.NET implementation Experiments Conclusions References
Goal: Activity Recognition in Smart Homes
Noisy data
−→ methods that characterise the noise
Tom Diethe Amazon Research Cambridge; University of Bristol
Activity Recognition from Accelerometers in Smart Homes
Introduction Methods Bayesian Dictionary Learning Infer.NET implementation Experiments Conclusions References
Goal: Activity Recognition in Smart Homes
Noisy data
−→ methods that characterise the noise
Missing packets
−→ robust methods
Tom Diethe Amazon Research Cambridge; University of Bristol
Activity Recognition from Accelerometers in Smart Homes
Introduction Methods Bayesian Dictionary Learning Infer.NET implementation Experiments Conclusions References
Goal: Activity Recognition in Smart Homes
Noisy data
−→ methods that characterise the noise
Missing packets
−→ robust methods
Limited storage and transmission capacity
−→ sparse/compressive methods
Tom Diethe Amazon Research Cambridge; University of Bristol
Activity Recognition from Accelerometers in Smart Homes
Introduction Methods Bayesian Dictionary Learning Infer.NET implementation Experiments Conclusions References
Goal: Activity Recognition in Smart Homes
Noisy data
−→ methods that characterise the noise
Missing packets
−→ robust methods
Limited storage and transmission capacity
−→ sparse/compressive methods
Labelled data time-consuming and costly to acquire
Tom Diethe Amazon Research Cambridge; University of Bristol
Activity Recognition from Accelerometers in Smart Homes
Introduction Methods Bayesian Dictionary Learning Infer.NET implementation Experiments Conclusions References
Goal: Activity Recognition in Smart Homes
Noisy data
−→ methods that characterise the noise
Missing packets
−→ robust methods
Limited storage and transmission capacity
−→ sparse/compressive methods
Labelled data time-consuming and costly to acquire
−→ Active Learning
Tom Diethe Amazon Research Cambridge; University of Bristol
Activity Recognition from Accelerometers in Smart Homes
Introduction Methods Bayesian Dictionary Learning Infer.NET implementation Experiments Conclusions References
Goal: Activity Recognition in Smart Homes
Noisy data
−→ methods that characterise the noise
Missing packets
−→ robust methods
Limited storage and transmission capacity
−→ sparse/compressive methods
Labelled data time-consuming and costly to acquire
−→ Active Learning
Differing training and deployment settings
Tom Diethe Amazon Research Cambridge; University of Bristol
Activity Recognition from Accelerometers in Smart Homes
Introduction Methods Bayesian Dictionary Learning Infer.NET implementation Experiments Conclusions References
Goal: Activity Recognition in Smart Homes
Noisy data
−→ methods that characterise the noise
Missing packets
−→ robust methods
Limited storage and transmission capacity
−→ sparse/compressive methods
Labelled data time-consuming and costly to acquire
−→ Active Learning
Differing training and deployment settings
−→ Transfer Learning
Tom Diethe Amazon Research Cambridge; University of Bristol
Activity Recognition from Accelerometers in Smart Homes
Introduction Methods Bayesian Dictionary Learning Infer.NET implementation Experiments Conclusions References
Motivation
Tom Diethe Amazon Research Cambridge; University of Bristol
Activity Recognition from Accelerometers in Smart Homes
Introduction Methods Bayesian Dictionary Learning Infer.NET implementation Experiments Conclusions References
Motivation
Would like a sparse
representation of the data
Tom Diethe Amazon Research Cambridge; University of Bristol
Activity Recognition from Accelerometers in Smart Homes
Introduction Methods Bayesian Dictionary Learning Infer.NET implementation Experiments Conclusions References
Motivation
Would like a sparse
representation of the data
Don’t want to make
parametric assumptions
Tom Diethe Amazon Research Cambridge; University of Bristol
Activity Recognition from Accelerometers in Smart Homes
Introduction Methods Bayesian Dictionary Learning Infer.NET implementation Experiments Conclusions References
Motivation
Would like a sparse
representation of the data
Don’t want to make
parametric assumptions
Features for classification
(activity recognition)
Tom Diethe Amazon Research Cambridge; University of Bristol
Activity Recognition from Accelerometers in Smart Homes
Introduction Methods Bayesian Dictionary Learning Infer.NET implementation Experiments Conclusions References
Motivation
Would like a sparse
representation of the data
Don’t want to make
parametric assumptions
Features for classification
(activity recognition)
−→ Dictionary Learning
Tom Diethe Amazon Research Cambridge; University of Bristol
Activity Recognition from Accelerometers in Smart Homes
Introduction Methods Bayesian Dictionary Learning Infer.NET implementation Experiments Conclusions References
Motivation
Would like a sparse
representation of the data
Don’t want to make
parametric assumptions
Features for classification
(activity recognition)
−→ Dictionary Learning
Would like to incorporate
with other models
Tom Diethe Amazon Research Cambridge; University of Bristol
Activity Recognition from Accelerometers in Smart Homes
Introduction Methods Bayesian Dictionary Learning Infer.NET implementation Experiments Conclusions References
Motivation
Would like a sparse
representation of the data
Don’t want to make
parametric assumptions
Features for classification
(activity recognition)
−→ Dictionary Learning
Would like to incorporate
with other models
−→ Bayesian approach
Tom Diethe Amazon Research Cambridge; University of Bristol
Activity Recognition from Accelerometers in Smart Homes
Introduction Methods Bayesian Dictionary Learning Infer.NET implementation Experiments Conclusions References
Dictionary Learning
Aim: find a set of vectors di , (dictionary), to represent
x ∈ Rn as a linear combination of these vectors:
x =
k
i=1
zi di s.t. k n. (1)
Tom Diethe Amazon Research Cambridge; University of Bristol
Activity Recognition from Accelerometers in Smart Homes
Introduction Methods Bayesian Dictionary Learning Infer.NET implementation Experiments Conclusions References
Dictionary Learning
Aim: find a set of vectors di , (dictionary), to represent
x ∈ Rn as a linear combination of these vectors:
x =
k
i=1
zi di s.t. k n. (1)
Sparsity: few non-zero components zi (or many close to 0)
Tom Diethe Amazon Research Cambridge; University of Bristol
Activity Recognition from Accelerometers in Smart Homes
Introduction Methods Bayesian Dictionary Learning Infer.NET implementation Experiments Conclusions References
Dictionary Learning
Aim: find a set of vectors di , (dictionary), to represent
x ∈ Rn as a linear combination of these vectors:
x =
k
i=1
zi di s.t. k n. (1)
Sparsity: few non-zero components zi (or many close to 0)
Cost function for m input vectors arranged in columns of
matrix X ∈ Rn×m
min
Z,D
X − DZ 2
F + λ
n
i=1
Ω(zi )
s.t. di
2
≤ C, ∀i = 1, . . . , k.
where D ∈ Rn×k is the dictionary, Z ∈ Rk×m are the
coefficients, and Ω(·) is a sparsity inducing regulariser, and λ
determines the relative importance of good reconstructionsTom Diethe Amazon Research Cambridge; University of Bristol
Activity Recognition from Accelerometers in Smart Homes
Introduction Methods Bayesian Dictionary Learning Infer.NET implementation Experiments Conclusions References
Generative model for Equation (1)
x
N
dot
d z
0
0
⌧
a b
N
N
Ga(1, 1)
Ga
n m
k
Based on Yang et al. (2015).
Tom Diethe Amazon Research Cambridge; University of Bristol
Activity Recognition from Accelerometers in Smart Homes
Introduction Methods Bayesian Dictionary Learning Infer.NET implementation Experiments Conclusions References
Generative model for Equation (1)
x
N
dot
d z
0 0
⌧
a b
N N
Ga(1, 1)
Ga
Ga(1, 1)
n m
k
Based on Yang et al. (2015).
Gamma prior over β, which
allows the dictionary atoms
to be automatically scaled
Tom Diethe Amazon Research Cambridge; University of Bristol
Activity Recognition from Accelerometers in Smart Homes
Introduction Methods Bayesian Dictionary Learning Infer.NET implementation Experiments Conclusions References
Generative model for Equation (1)
x
N
dot
d z
↵
0
⌧
a b
N N
Ga(1, 1)
Ga
N(0, 1) Ga(1, 1)
n m
k
Based on Yang et al. (2015).
Gamma prior over β, which
allows the dictionary atoms
to be automatically scaled
additional level of hierarchy
through the variables α on
the means of the dictionary
components, which can aid
online learning
Tom Diethe Amazon Research Cambridge; University of Bristol
Activity Recognition from Accelerometers in Smart Homes
Introduction Methods Bayesian Dictionary Learning Infer.NET implementation Experiments Conclusions References
Generative model for Equation (1)
x
N
dot
d z
↵
0
⌧
a b
N N
Ga(1, 1)
Ga
N(0, 1) Ga(1, 1)
n m
k
Based on Yang et al. (2015).
Gamma prior over β, which
allows the dictionary atoms
to be automatically scaled
additional level of hierarchy
through the variables α on
the means of the dictionary
components, which can aid
online learning
a, b control sparsity:
Ga(1, 1) → non-sparse
Ga(0.5, 10−
6) → sparse
Tom Diethe Amazon Research Cambridge; University of Bristol
Activity Recognition from Accelerometers in Smart Homes
Introduction Methods Bayesian Dictionary Learning Infer.NET implementation Experiments Conclusions References
Infer.NET
Mode (.NET code)
Infer.NET Compiler
Inference code
Answers!Data
Framework for running
Bayesian inference in
graphical modelsa
Experiments used Monob,
running on OS-X and Linux
Inference engine: Variational
Message Passing (VMP),
deterministic approximation
algorithm
a
http://research.microsoft.com/infernet
b
http://www.mono-project.com/
Tom Diethe Amazon Research Cambridge; University of Bristol
Activity Recognition from Accelerometers in Smart Homes
Introduction Methods Bayesian Dictionary Learning Infer.NET implementation Experiments Conclusions References
Modelling code
p u b l i c void BDL( double [ , ] x , i n t numBases , double a , double b )
{
var b a s i s = new Range ( numBases ) ;
var s i g n a l = new Range ( x . GetLength ( 0 ) ) ;
var sample = new Range ( x . GetLength ( 1 ) ) ;
var n o i s e P r e c i s i o n = V a r i a b l e . GammaFromShapeAndRate (1 , 1 ) ;
var c o e f f i c i e n t P r e c i s i o n s = V a r i a b l e . Array<double >(s i g n a l , b a s i s ) ;
var dictionaryMeans = V a r i a b l e . Array<double >(b a s i s , sample ) ;
var d i c t i o n a r y P r e c i s i o n s = V a r i a b l e . Array<double >(b a s i s , sample ) ;
var c o e f f i c i e n t s = V a r i a b l e . Array<double >(s i g n a l , b a s i s ) ;
var d i c t i o n a r y = V a r i a b l e . Array<double >(b a s i s , sample ) ;
var s i g n a l s = V a r i a b l e . Array<double >(s i g n a l , sample ) ;
dictionaryMeans [ b a s i s ] [ sample ] = V a r i a b l e . GaussianFromMeanAndVariance (0 , 1 ) . ForEach ( b a
d i c t i o n a r y P r e c i s i o n s [ b a s i s , sample ] = V a r i a b l e . GammaFromShapeAndRate (1 , 1 ) . ForEach ( b a s
c o e f f i c i e n t P r e c i s i o n s [ s i g n a l , b a s i s ] = V a r i a b l e . GammaFromShapeAndRate ( a , b ) . ForEach ( s i
c o e f f i c i e n t s [ s i g n a l , b a s i s ] = V a r i a b l e . GaussianFromMeanAndPrecision (0 , c o e f f i c i e n t P r e c
d i c t i o n a r y [ b a s i s , sample ] = V a r i a b l e . GaussianFromMeanAndPrecision (
dictionaryMeans [ b a s i s , sample ] , d i c t i o n a r y P r e c i s i o n s [ b a s i s , sample ] ) ;
d i c t i o n a r y [ b a s i s , sample ] . I n i t i a l i s e T o ( d i c t i o n a r y P r i o r s [ b a s i s , sample ] ) ;
var c l e a n S i g n a l s = V a r i a b l e . M a t r i x M u l t i p l y ( c o e f f i c i e n t s , d i c t i o n a r y ) . Named( ” c l e a n ” ) ;
s i g n a l s [ s i g n a l , sample ] = V a r i a b l e . GaussianFromMeanAndPrecision ( c l e a n S i g n a l s [ s i g n a l , s
s i g n a l s . ObservedValue = x ;
}
Tom Diethe Amazon Research Cambridge; University of Bristol
Activity Recognition from Accelerometers in Smart Homes
Introduction Methods Bayesian Dictionary Learning Infer.NET implementation Experiments Conclusions References
Datasets
HAR dataset Anguita et al. (2013)
Publicly available activity recognition dataset1
Smart-phone on the waist, with tri-axial accelerometer, 50 Hz
Annotation was done using video-recordings.
Activities: Lie, Stand, Walk, Sit, Ascend stairs, Descend stairs
SPHERE challenge dataset Twomey et al. (2016)
collected by SPHERE project and made public as a challenge2
Tri-axial accelerometer on dominant wrist, 20 Hz, range ±8 g
Activites: Lie, Stand, Walk
1
https://archive.ics.uci.edu/ml/datasets/Human+Activity+
Recognition+Using+Smartphones
2
http://irc-sphere.ac.uk/sphere-challenge/home
Tom Diethe Amazon Research Cambridge; University of Bristol
Activity Recognition from Accelerometers in Smart Homes
Introduction Methods Bayesian Dictionary Learning Infer.NET implementation Experiments Conclusions References
Sparsity
Non-sparse priors (Ga(1, 1))
average sparsity 0.84
bases
signals
Sparse priors (Ga(0.5, 10−6))
average sparsity 0.96
bases
signals
Tom Diethe Amazon Research Cambridge; University of Bristol
Activity Recognition from Accelerometers in Smart Homes
Introduction Methods Bayesian Dictionary Learning Infer.NET implementation Experiments Conclusions References
Convergence of VMP
Model evidence converges monotonically to a local maximum
0 2 4 6 8 10 12 14 16 18
Iterations
350000
300000
250000
200000
150000
100000
50000
0
50000
Evidencelogodds
non-sparse
sparse
Tom Diethe Amazon Research Cambridge; University of Bristol
Activity Recognition from Accelerometers in Smart Homes
Introduction Methods Bayesian Dictionary Learning Infer.NET implementation Experiments Conclusions References
Convergence of reconstruction error
0 2 4 6 8 10 12 14 16 18
Iterations
0.02
0.03
0.04
0.05
0.06
0.07
0.08
0.09
RMSE
non-sparse
sparse
Tom Diethe Amazon Research Cambridge; University of Bristol
Activity Recognition from Accelerometers in Smart Homes
Introduction Methods Bayesian Dictionary Learning Infer.NET implementation Experiments Conclusions References
Learnt Dictionary
150000
100000
50000
0
50000
100000
150000
150000
100000
50000
0
50000
100000
150000
150000
100000
50000
0
50000
100000
150000
20020406080100120140160
150000
100000
50000
0
50000
100000
150000
20020406080100120140160 20020406080100120140160 20020406080100120140160
Dictionary 0-16
x
y
Figure: 16 example bases from the dictionary of 128 bases inferred by BDL using the
non-sparse priors on HAR dataset
Tom Diethe Amazon Research Cambridge; University of Bristol
Activity Recognition from Accelerometers in Smart Homes
Introduction Methods Bayesian Dictionary Learning Infer.NET implementation Experiments Conclusions References
Coefficients
Tom Diethe Amazon Research Cambridge; University of Bristol
Activity Recognition from Accelerometers in Smart Homes
Introduction Methods Bayesian Dictionary Learning Infer.NET implementation Experiments Conclusions References
Reconstructions
0.6
0.4
0.2
0.0
0.2
0.4
0.6
0.8
0.0 0.5 1.0 1.5 2.0 2.5 3.0
0.6
0.4
0.2
0.0
0.2
0.4
0.6
0.8
signal reconstruction ±s.d.
Reconstructions, RMSE=0.0276
time (s)
magnitude
Tom Diethe Amazon Research Cambridge; University of Bristol
Activity Recognition from Accelerometers in Smart Homes
Introduction Methods Bayesian Dictionary Learning Infer.NET implementation Experiments Conclusions References
Reconstruction Performance
SPAMS BDL Sparse BDL
Bases RMSE Sparsity RMSE Sparsity RMSE Sparsity
64 0.0480 0.71 0.0519 0.88 0.0293 0.62
128 0.0457 0.84 0.0400 0.94 0.0276 0.84
256 0.0438 0.92 0.0316 0.99 0.0288 0.93
512 0.0423 0.96 0.0224 0.99 0.0231 0.96
Tom Diethe Amazon Research Cambridge; University of Bristol
Activity Recognition from Accelerometers in Smart Homes
Introduction Methods Bayesian Dictionary Learning Infer.NET implementation Experiments Conclusions References
Classification
Multi-class Bayes Point Machine (BPM)
Herbrich et al. (2001), also using
Infer.NET
Assumptions:
Feature values x are always fully
observed
The order of instances does not matter
The predictive distribution is a linear
discriminant of the form:
p(yi |xi , w) = p(yi |si = w xi ), where w
are the weights and si is the score for
instance i
y
˜s
s
1
N
w x
⇥
µw ⌧w
N
N
D
A
R
Tom Diethe Amazon Research Cambridge; University of Bristol
Activity Recognition from Accelerometers in Smart Homes
Introduction Methods Bayesian Dictionary Learning Infer.NET implementation Experiments Conclusions References
Activity Recognition Results
Use coefficients as features of classification algorithm
64 bases + bias feature
Metric: per-class 1 vs rest area under ROC curve
HAR dataset
Activity BDL sparse BDL SPAMS
Walking 0.73 0.83 0.88
Ascending stairs 0.63 0.60 0.83
Descending stairs 0.61 0.34 0.82
Sitting 0.74 0.72 0.89
Standing 0.51 0.43 0.98
Lying down 0.95 0.95 0.95
Average 0.70 0.65 0.89
SPHERE dataset
Activity BDL sparse BDL SPAMS
Walking 0.55 0.51 0.46
Standing 0.69 0.62 0.44
Lying down 0.86 0.85 0.46
Average 0.70 0.66 0.45
Tom Diethe Amazon Research Cambridge; University of Bristol
Activity Recognition from Accelerometers in Smart Homes
Introduction Methods Bayesian Dictionary Learning Infer.NET implementation Experiments Conclusions References
Transfer Learning
Source: DS = {(xi , yi )mS
i=1}
Target: DT = {(xi , yi )mT
i=1}
Assumptions:
labels available for source domain
labels can be acquired for target domain, but costly
Tom Diethe Amazon Research Cambridge; University of Bristol
Activity Recognition from Accelerometers in Smart Homes
Introduction Methods Bayesian Dictionary Learning Infer.NET implementation Experiments Conclusions References
Hierarchical Model
Multiclass Bayes Point
Machine (BPM)
y
˜s
s
1
N
w x
⇥
µw ⌧w
N
N
D
A
R
Tom Diethe Amazon Research Cambridge; University of Bristol
Activity Recognition from Accelerometers in Smart Homes
Introduction Methods Bayesian Dictionary Learning Infer.NET implementation Experiments Conclusions References
Hierarchical Model
Community: Learn
community posteriors
y
˜s
s
1
N
w x
⇥
µw ⌧w
µµ µ k⌧ ✓⌧
N
N
N
D
A
R
Tom Diethe Amazon Research Cambridge; University of Bristol
Activity Recognition from Accelerometers in Smart Homes
Introduction Methods Bayesian Dictionary Learning Infer.NET implementation Experiments Conclusions References
Hierarchical Model
Transfer: weight means
and precisions replaced by
community posteriors from
source
y
˜s
s
1
N
w x
⇥
µw ⌧w
N
N
D
A
R
Tom Diethe Amazon Research Cambridge; University of Bristol
Activity Recognition from Accelerometers in Smart Homes
Introduction Methods Bayesian Dictionary Learning Infer.NET implementation Experiments Conclusions References
Hierarchical Model
Personalisation:
smoothly evolve from
generic to custom
predictions
y
˜s
s
1
N
w x
⇥
µw ⌧w
N
N
D
A
R
Tom Diethe Amazon Research Cambridge; University of Bristol
Activity Recognition from Accelerometers in Smart Homes
Introduction Methods Bayesian Dictionary Learning Infer.NET implementation Experiments Conclusions References
Active Learning
Baseline: random (= online learning)
Uncertainty Sampling
Nearest to decision boundary, fails in case of corrupted data
Certainty Sampling
Furthest from decision boundary. Normally nonsensical but
solves corruption issue
Value of Information (VOI) Kapoor et al. (2007)
Decision theoretic measure that uses p(y = 1) vs p(y = 0)
(relative risk). Expensive to compute.
Tom Diethe Amazon Research Cambridge; University of Bristol
Activity Recognition from Accelerometers in Smart Homes
Introduction Methods Bayesian Dictionary Learning Infer.NET implementation Experiments Conclusions References
Datasets
Source: Anguita et al. (2013) 30 subjects, Smartphone, 50Hz,
Video annotations
Target: Zhang and Sawchuk (2012) 14 subjects, MotionNode,
100Hz, Observer annotations
Classes: Walking upstairs vs. Walking downstairs
Tom Diethe Amazon Research Cambridge; University of Bristol
Activity Recognition from Accelerometers in Smart Homes
Introduction Methods Bayesian Dictionary Learning Infer.NET implementation Experiments Conclusions References
Results
0 5 10 15 20
# instances
0.0
0.2
0.4
0.6
0.8
1.0Accuracy Hold-out Accuracy
Random
US
CS
VOI+
VOI-
Tom Diethe Amazon Research Cambridge; University of Bristol
Activity Recognition from Accelerometers in Smart Homes
Introduction Methods Bayesian Dictionary Learning Infer.NET implementation Experiments Conclusions References
Conclusions and Further Work
Improved methods for Bayesian Dictionary Learning
Sparsity does not need to be enforced
Application to accelerometer signals for Activity Recognition
Hierarchical extension to BPM for transfer learning
Combination of Active & Transfer learning
−→ fast learning on target dataset
Uncertainty Sampling good enough
Tom Diethe Amazon Research Cambridge; University of Bristol
Activity Recognition from Accelerometers in Smart Homes
Introduction Methods Bayesian Dictionary Learning Infer.NET implementation Experiments Conclusions References
Conclusions and Further Work
Improved methods for Bayesian Dictionary Learning
Sparsity does not need to be enforced
Application to accelerometer signals for Activity Recognition
Hierarchical extension to BPM for transfer learning
Combination of Active & Transfer learning
−→ fast learning on target dataset
Uncertainty Sampling good enough
Further Work:
Incorporate classifier directly into the model
Tom Diethe Amazon Research Cambridge; University of Bristol
Activity Recognition from Accelerometers in Smart Homes
Introduction Methods Bayesian Dictionary Learning Infer.NET implementation Experiments Conclusions References
Conclusions and Further Work
Improved methods for Bayesian Dictionary Learning
Sparsity does not need to be enforced
Application to accelerometer signals for Activity Recognition
Hierarchical extension to BPM for transfer learning
Combination of Active & Transfer learning
−→ fast learning on target dataset
Uncertainty Sampling good enough
Further Work:
Incorporate classifier directly into the model
Source code:
https://github.com/IRC-SPHERE/bayesian-dictionary-learning
https://github.com/IRC-SPHERE/ActiveTransfer
Tom Diethe Amazon Research Cambridge; University of Bristol
Activity Recognition from Accelerometers in Smart Homes
Introduction Methods Bayesian Dictionary Learning Infer.NET implementation Experiments Conclusions References
Selected References
D Anguita, A Ghio, L Oneto, X Parra, and JL Reyes-Ortiz. A public domain dataset for human activity recognition
using smartphones. In ESANN, 2013.
Tom Diethe, Niall Twomey, and Peter Flach. Active transfer learning for activity recognition. In 24th European
Symposium on Artificial Neural Networks, ESANN 2016, Bruges, Belgium, April 27-29, 2016, 2016a.
Tom Diethe, Niall Twomey, and Peter Flach. BDL. NET: Bayesian dictionary learning in Infer. NET. In Machine
Learning for Signal Processing (MLSP), 2016 IEEE 26th International Workshop on, pages 1–6. IEEE, 2016b.
Ralf Herbrich, Thore Graepel, and Colin Campbell. Bayes point machines. Journal of Machine Learning Research,
1:245–279, January 2001. URL http://research.microsoft.com/apps/pubs/default.aspx?id=65611.
Ashish Kapoor, Eric Horvitz, and Sumit Basu. Selective supervision: Guiding supervised learning with
decision-theoretic active learning. In IJCAI, volume 7, pages 877–882, 2007.
Niall Twomey, Tom Diethe, Meelis Kull, Hao Song, Massimo Camplani, Sion Hannuna, Xenofon Fafoutis, Ni Zhu,
Pete Woznowski, Peter Flach, and Ian Craddock. The sphere challenge: Activity recognition with multimodal
sensor data. arXiv preprint arXiv:1603.00797, 2016.
Linxiao Yang, Jun Fang, Hong Cheng, and Hongbin Li. Sparse Bayesian dictionary learning with a Gaussian
hierarchical model. CoRR, abs/1503.02144, 2015. URL http://arxiv.org/abs/1503.02144.
M. Zhang and A.A. Sawchuk. USC-HAD: A daily activity dataset for ubiquitous activity recognition using wearable
sensors. In ACM Int. Conf. on Ubiquitous Computing Workshop on Situation, Activity and Goal Awareness
(SAGAware), 2012.
Tom Diethe Amazon Research Cambridge; University of Bristol
Activity Recognition from Accelerometers in Smart Homes

Mais conteúdo relacionado

Mais procurados

5 Lessons Learned from Designing Neural Models for Information Retrieval
5 Lessons Learned from Designing Neural Models for Information Retrieval5 Lessons Learned from Designing Neural Models for Information Retrieval
5 Lessons Learned from Designing Neural Models for Information RetrievalBhaskar Mitra
 
Predictive Text Embedding using LINE
Predictive Text Embedding using LINEPredictive Text Embedding using LINE
Predictive Text Embedding using LINENishant Prateek
 
Conformer-Kernel with Query Term Independence @ TREC 2020 Deep Learning Track
Conformer-Kernel with Query Term Independence @ TREC 2020 Deep Learning TrackConformer-Kernel with Query Term Independence @ TREC 2020 Deep Learning Track
Conformer-Kernel with Query Term Independence @ TREC 2020 Deep Learning TrackBhaskar Mitra
 
Modeling documents with Generative Adversarial Networks - John Glover
Modeling documents with Generative Adversarial Networks - John GloverModeling documents with Generative Adversarial Networks - John Glover
Modeling documents with Generative Adversarial Networks - John GloverSebastian Ruder
 
Reverse engineering and theory building v3
Reverse engineering and theory building v3Reverse engineering and theory building v3
Reverse engineering and theory building v3ClarkTony
 
Joint Word and Entity Embeddings for Entity Retrieval from Knowledge Graph
Joint Word and Entity Embeddings for Entity Retrieval from Knowledge GraphJoint Word and Entity Embeddings for Entity Retrieval from Knowledge Graph
Joint Word and Entity Embeddings for Entity Retrieval from Knowledge GraphFedorNikolaev
 
집합모델 확장불린모델
집합모델  확장불린모델집합모델  확장불린모델
집합모델 확장불린모델guesta34d441
 
Using Text Embeddings for Information Retrieval
Using Text Embeddings for Information RetrievalUsing Text Embeddings for Information Retrieval
Using Text Embeddings for Information RetrievalBhaskar Mitra
 
Neural Models for Document Ranking
Neural Models for Document RankingNeural Models for Document Ranking
Neural Models for Document RankingBhaskar Mitra
 
Dual Embedding Space Model (DESM)
Dual Embedding Space Model (DESM)Dual Embedding Space Model (DESM)
Dual Embedding Space Model (DESM)Bhaskar Mitra
 
Neural Text Embeddings for Information Retrieval (WSDM 2017)
Neural Text Embeddings for Information Retrieval (WSDM 2017)Neural Text Embeddings for Information Retrieval (WSDM 2017)
Neural Text Embeddings for Information Retrieval (WSDM 2017)Bhaskar Mitra
 
Duet @ TREC 2019 Deep Learning Track
Duet @ TREC 2019 Deep Learning TrackDuet @ TREC 2019 Deep Learning Track
Duet @ TREC 2019 Deep Learning TrackBhaskar Mitra
 
Bytewise Approximate Match: Theory, Algorithms and Applications
Bytewise Approximate Match:  Theory, Algorithms and ApplicationsBytewise Approximate Match:  Theory, Algorithms and Applications
Bytewise Approximate Match: Theory, Algorithms and ApplicationsLiwei Ren任力偉
 
Deep Learning for Search
Deep Learning for SearchDeep Learning for Search
Deep Learning for SearchBhaskar Mitra
 
Basic review on topic modeling
Basic review on  topic modelingBasic review on  topic modeling
Basic review on topic modelingHiroyuki Kuromiya
 
From Sensing to Decision
From Sensing to DecisionFrom Sensing to Decision
From Sensing to DecisionTzar Umang
 
Natural Language Processing for Materials Design - What Can We Extract From t...
Natural Language Processing for Materials Design - What Can We Extract From t...Natural Language Processing for Materials Design - What Can We Extract From t...
Natural Language Processing for Materials Design - What Can We Extract From t...Anubhav Jain
 

Mais procurados (20)

5 Lessons Learned from Designing Neural Models for Information Retrieval
5 Lessons Learned from Designing Neural Models for Information Retrieval5 Lessons Learned from Designing Neural Models for Information Retrieval
5 Lessons Learned from Designing Neural Models for Information Retrieval
 
Predictive Text Embedding using LINE
Predictive Text Embedding using LINEPredictive Text Embedding using LINE
Predictive Text Embedding using LINE
 
Conformer-Kernel with Query Term Independence @ TREC 2020 Deep Learning Track
Conformer-Kernel with Query Term Independence @ TREC 2020 Deep Learning TrackConformer-Kernel with Query Term Independence @ TREC 2020 Deep Learning Track
Conformer-Kernel with Query Term Independence @ TREC 2020 Deep Learning Track
 
Video+Language: From Classification to Description
Video+Language: From Classification to DescriptionVideo+Language: From Classification to Description
Video+Language: From Classification to Description
 
Modeling documents with Generative Adversarial Networks - John Glover
Modeling documents with Generative Adversarial Networks - John GloverModeling documents with Generative Adversarial Networks - John Glover
Modeling documents with Generative Adversarial Networks - John Glover
 
Reverse engineering and theory building v3
Reverse engineering and theory building v3Reverse engineering and theory building v3
Reverse engineering and theory building v3
 
Joint Word and Entity Embeddings for Entity Retrieval from Knowledge Graph
Joint Word and Entity Embeddings for Entity Retrieval from Knowledge GraphJoint Word and Entity Embeddings for Entity Retrieval from Knowledge Graph
Joint Word and Entity Embeddings for Entity Retrieval from Knowledge Graph
 
집합모델 확장불린모델
집합모델  확장불린모델집합모델  확장불린모델
집합모델 확장불린모델
 
Using Text Embeddings for Information Retrieval
Using Text Embeddings for Information RetrievalUsing Text Embeddings for Information Retrieval
Using Text Embeddings for Information Retrieval
 
Understanding GloVe
Understanding GloVeUnderstanding GloVe
Understanding GloVe
 
Neural Models for Document Ranking
Neural Models for Document RankingNeural Models for Document Ranking
Neural Models for Document Ranking
 
Dual Embedding Space Model (DESM)
Dual Embedding Space Model (DESM)Dual Embedding Space Model (DESM)
Dual Embedding Space Model (DESM)
 
Neural Text Embeddings for Information Retrieval (WSDM 2017)
Neural Text Embeddings for Information Retrieval (WSDM 2017)Neural Text Embeddings for Information Retrieval (WSDM 2017)
Neural Text Embeddings for Information Retrieval (WSDM 2017)
 
Duet @ TREC 2019 Deep Learning Track
Duet @ TREC 2019 Deep Learning TrackDuet @ TREC 2019 Deep Learning Track
Duet @ TREC 2019 Deep Learning Track
 
Bytewise Approximate Match: Theory, Algorithms and Applications
Bytewise Approximate Match:  Theory, Algorithms and ApplicationsBytewise Approximate Match:  Theory, Algorithms and Applications
Bytewise Approximate Match: Theory, Algorithms and Applications
 
Deep Learning for Search
Deep Learning for SearchDeep Learning for Search
Deep Learning for Search
 
Video google-text (2)
Video google-text (2)Video google-text (2)
Video google-text (2)
 
Basic review on topic modeling
Basic review on  topic modelingBasic review on  topic modeling
Basic review on topic modeling
 
From Sensing to Decision
From Sensing to DecisionFrom Sensing to Decision
From Sensing to Decision
 
Natural Language Processing for Materials Design - What Can We Extract From t...
Natural Language Processing for Materials Design - What Can We Extract From t...Natural Language Processing for Materials Design - What Can We Extract From t...
Natural Language Processing for Materials Design - What Can We Extract From t...
 

Semelhante a Activity Recognition from Accelerometers in Smart Homes

IRJET- Big Data and Bayes Theorem used Analyze the Student’s Performance in E...
IRJET- Big Data and Bayes Theorem used Analyze the Student’s Performance in E...IRJET- Big Data and Bayes Theorem used Analyze the Student’s Performance in E...
IRJET- Big Data and Bayes Theorem used Analyze the Student’s Performance in E...IRJET Journal
 
32_Nov07_MachineLear..
32_Nov07_MachineLear..32_Nov07_MachineLear..
32_Nov07_MachineLear..butest
 
Clustering
ClusteringClustering
Clusteringbutest
 
SHAHBAZ_TECHNICAL_SEMINAR.docx
SHAHBAZ_TECHNICAL_SEMINAR.docxSHAHBAZ_TECHNICAL_SEMINAR.docx
SHAHBAZ_TECHNICAL_SEMINAR.docxShahbazKhan77289
 
Social Network Analysis: What It Is, Why We Should Care, and What We Can Lear...
Social Network Analysis: What It Is, Why We Should Care, and What We Can Lear...Social Network Analysis: What It Is, Why We Should Care, and What We Can Lear...
Social Network Analysis: What It Is, Why We Should Care, and What We Can Lear...Xiaohan Zeng
 
kantorNSF-NIJ-ISI-03-06-04.ppt
kantorNSF-NIJ-ISI-03-06-04.pptkantorNSF-NIJ-ISI-03-06-04.ppt
kantorNSF-NIJ-ISI-03-06-04.pptbutest
 
Tutorial - Recommender systems meet linked open data - ICWE 2016 - Lugano - 0...
Tutorial - Recommender systems meet linked open data - ICWE 2016 - Lugano - 0...Tutorial - Recommender systems meet linked open data - ICWE 2016 - Lugano - 0...
Tutorial - Recommender systems meet linked open data - ICWE 2016 - Lugano - 0...Polytechnic University of Bari
 
Big Data + Big Sim: Query Processing over Unstructured CFD Models
Big Data + Big Sim: Query Processing over Unstructured CFD ModelsBig Data + Big Sim: Query Processing over Unstructured CFD Models
Big Data + Big Sim: Query Processing over Unstructured CFD ModelsUniversity of Washington
 
probabilistic ranking
probabilistic rankingprobabilistic ranking
probabilistic rankingFELIX75
 
A Distributional Semantics Approach for Selective Reasoning on Commonsense Gr...
A Distributional Semantics Approach for Selective Reasoning on Commonsense Gr...A Distributional Semantics Approach for Selective Reasoning on Commonsense Gr...
A Distributional Semantics Approach for Selective Reasoning on Commonsense Gr...Andre Freitas
 
AutoML for Data Science Productivity and Toward Better Digital Decisions
AutoML for Data Science Productivity and Toward Better Digital DecisionsAutoML for Data Science Productivity and Toward Better Digital Decisions
AutoML for Data Science Productivity and Toward Better Digital DecisionsSteven Gustafson
 
Computer notes - data structures
Computer notes - data structuresComputer notes - data structures
Computer notes - data structuresecomputernotes
 
EE-232-LEC-01 Data_structures.pptx
EE-232-LEC-01 Data_structures.pptxEE-232-LEC-01 Data_structures.pptx
EE-232-LEC-01 Data_structures.pptxiamultapromax
 
A study on rough set theory based
A study on rough set theory basedA study on rough set theory based
A study on rough set theory basedijaia
 
Data Tactics Data Science Brown Bag (April 2014)
Data Tactics Data Science Brown Bag (April 2014)Data Tactics Data Science Brown Bag (April 2014)
Data Tactics Data Science Brown Bag (April 2014)Rich Heimann
 
Data Science and Analytics Brown Bag
Data Science and Analytics Brown BagData Science and Analytics Brown Bag
Data Science and Analytics Brown BagDataTactics
 

Semelhante a Activity Recognition from Accelerometers in Smart Homes (20)

IRJET- Big Data and Bayes Theorem used Analyze the Student’s Performance in E...
IRJET- Big Data and Bayes Theorem used Analyze the Student’s Performance in E...IRJET- Big Data and Bayes Theorem used Analyze the Student’s Performance in E...
IRJET- Big Data and Bayes Theorem used Analyze the Student’s Performance in E...
 
fINAL ML PPT.pptx
fINAL ML PPT.pptxfINAL ML PPT.pptx
fINAL ML PPT.pptx
 
32_Nov07_MachineLear..
32_Nov07_MachineLear..32_Nov07_MachineLear..
32_Nov07_MachineLear..
 
Clustering
ClusteringClustering
Clustering
 
Density based clustering
Density based clusteringDensity based clustering
Density based clustering
 
SHAHBAZ_TECHNICAL_SEMINAR.docx
SHAHBAZ_TECHNICAL_SEMINAR.docxSHAHBAZ_TECHNICAL_SEMINAR.docx
SHAHBAZ_TECHNICAL_SEMINAR.docx
 
Social Network Analysis: What It Is, Why We Should Care, and What We Can Lear...
Social Network Analysis: What It Is, Why We Should Care, and What We Can Lear...Social Network Analysis: What It Is, Why We Should Care, and What We Can Lear...
Social Network Analysis: What It Is, Why We Should Care, and What We Can Lear...
 
kantorNSF-NIJ-ISI-03-06-04.ppt
kantorNSF-NIJ-ISI-03-06-04.pptkantorNSF-NIJ-ISI-03-06-04.ppt
kantorNSF-NIJ-ISI-03-06-04.ppt
 
Tutorial - Recommender systems meet linked open data - ICWE 2016 - Lugano - 0...
Tutorial - Recommender systems meet linked open data - ICWE 2016 - Lugano - 0...Tutorial - Recommender systems meet linked open data - ICWE 2016 - Lugano - 0...
Tutorial - Recommender systems meet linked open data - ICWE 2016 - Lugano - 0...
 
2-IJCSE-00536
2-IJCSE-005362-IJCSE-00536
2-IJCSE-00536
 
2-IJCSE-00536
2-IJCSE-005362-IJCSE-00536
2-IJCSE-00536
 
Big Data + Big Sim: Query Processing over Unstructured CFD Models
Big Data + Big Sim: Query Processing over Unstructured CFD ModelsBig Data + Big Sim: Query Processing over Unstructured CFD Models
Big Data + Big Sim: Query Processing over Unstructured CFD Models
 
probabilistic ranking
probabilistic rankingprobabilistic ranking
probabilistic ranking
 
A Distributional Semantics Approach for Selective Reasoning on Commonsense Gr...
A Distributional Semantics Approach for Selective Reasoning on Commonsense Gr...A Distributional Semantics Approach for Selective Reasoning on Commonsense Gr...
A Distributional Semantics Approach for Selective Reasoning on Commonsense Gr...
 
AutoML for Data Science Productivity and Toward Better Digital Decisions
AutoML for Data Science Productivity and Toward Better Digital DecisionsAutoML for Data Science Productivity and Toward Better Digital Decisions
AutoML for Data Science Productivity and Toward Better Digital Decisions
 
Computer notes - data structures
Computer notes - data structuresComputer notes - data structures
Computer notes - data structures
 
EE-232-LEC-01 Data_structures.pptx
EE-232-LEC-01 Data_structures.pptxEE-232-LEC-01 Data_structures.pptx
EE-232-LEC-01 Data_structures.pptx
 
A study on rough set theory based
A study on rough set theory basedA study on rough set theory based
A study on rough set theory based
 
Data Tactics Data Science Brown Bag (April 2014)
Data Tactics Data Science Brown Bag (April 2014)Data Tactics Data Science Brown Bag (April 2014)
Data Tactics Data Science Brown Bag (April 2014)
 
Data Science and Analytics Brown Bag
Data Science and Analytics Brown BagData Science and Analytics Brown Bag
Data Science and Analytics Brown Bag
 

Último

Pests of safflower_Binomics_Identification_Dr.UPR.pdf
Pests of safflower_Binomics_Identification_Dr.UPR.pdfPests of safflower_Binomics_Identification_Dr.UPR.pdf
Pests of safflower_Binomics_Identification_Dr.UPR.pdfPirithiRaju
 
User Guide: Capricorn FLX™ Weather Station
User Guide: Capricorn FLX™ Weather StationUser Guide: Capricorn FLX™ Weather Station
User Guide: Capricorn FLX™ Weather StationColumbia Weather Systems
 
Observational constraints on mergers creating magnetism in massive stars
Observational constraints on mergers creating magnetism in massive starsObservational constraints on mergers creating magnetism in massive stars
Observational constraints on mergers creating magnetism in massive starsSérgio Sacani
 
Thermodynamics ,types of system,formulae ,gibbs free energy .pptx
Thermodynamics ,types of system,formulae ,gibbs free energy .pptxThermodynamics ,types of system,formulae ,gibbs free energy .pptx
Thermodynamics ,types of system,formulae ,gibbs free energy .pptxuniversity
 
GENERAL PHYSICS 2 REFRACTION OF LIGHT SENIOR HIGH SCHOOL GENPHYS2.pptx
GENERAL PHYSICS 2 REFRACTION OF LIGHT SENIOR HIGH SCHOOL GENPHYS2.pptxGENERAL PHYSICS 2 REFRACTION OF LIGHT SENIOR HIGH SCHOOL GENPHYS2.pptx
GENERAL PHYSICS 2 REFRACTION OF LIGHT SENIOR HIGH SCHOOL GENPHYS2.pptxRitchAndruAgustin
 
FREE NURSING BUNDLE FOR NURSES.PDF by na
FREE NURSING BUNDLE FOR NURSES.PDF by naFREE NURSING BUNDLE FOR NURSES.PDF by na
FREE NURSING BUNDLE FOR NURSES.PDF by naJASISJULIANOELYNV
 
Call Girls in Majnu Ka Tilla Delhi 🔝9711014705🔝 Genuine
Call Girls in Majnu Ka Tilla Delhi 🔝9711014705🔝 GenuineCall Girls in Majnu Ka Tilla Delhi 🔝9711014705🔝 Genuine
Call Girls in Majnu Ka Tilla Delhi 🔝9711014705🔝 Genuinethapagita
 
Pests of Blackgram, greengram, cowpea_Dr.UPR.pdf
Pests of Blackgram, greengram, cowpea_Dr.UPR.pdfPests of Blackgram, greengram, cowpea_Dr.UPR.pdf
Pests of Blackgram, greengram, cowpea_Dr.UPR.pdfPirithiRaju
 
Fertilization: Sperm and the egg—collectively called the gametes—fuse togethe...
Fertilization: Sperm and the egg—collectively called the gametes—fuse togethe...Fertilization: Sperm and the egg—collectively called the gametes—fuse togethe...
Fertilization: Sperm and the egg—collectively called the gametes—fuse togethe...D. B. S. College Kanpur
 
(9818099198) Call Girls In Noida Sector 14 (NOIDA ESCORTS)
(9818099198) Call Girls In Noida Sector 14 (NOIDA ESCORTS)(9818099198) Call Girls In Noida Sector 14 (NOIDA ESCORTS)
(9818099198) Call Girls In Noida Sector 14 (NOIDA ESCORTS)riyaescorts54
 
basic entomology with insect anatomy and taxonomy
basic entomology with insect anatomy and taxonomybasic entomology with insect anatomy and taxonomy
basic entomology with insect anatomy and taxonomyDrAnita Sharma
 
Microteaching on terms used in filtration .Pharmaceutical Engineering
Microteaching on terms used in filtration .Pharmaceutical EngineeringMicroteaching on terms used in filtration .Pharmaceutical Engineering
Microteaching on terms used in filtration .Pharmaceutical EngineeringPrajakta Shinde
 
Topic 9- General Principles of International Law.pptx
Topic 9- General Principles of International Law.pptxTopic 9- General Principles of International Law.pptx
Topic 9- General Principles of International Law.pptxJorenAcuavera1
 
User Guide: Pulsar™ Weather Station (Columbia Weather Systems)
User Guide: Pulsar™ Weather Station (Columbia Weather Systems)User Guide: Pulsar™ Weather Station (Columbia Weather Systems)
User Guide: Pulsar™ Weather Station (Columbia Weather Systems)Columbia Weather Systems
 
THE ROLE OF PHARMACOGNOSY IN TRADITIONAL AND MODERN SYSTEM OF MEDICINE.pptx
THE ROLE OF PHARMACOGNOSY IN TRADITIONAL AND MODERN SYSTEM OF MEDICINE.pptxTHE ROLE OF PHARMACOGNOSY IN TRADITIONAL AND MODERN SYSTEM OF MEDICINE.pptx
THE ROLE OF PHARMACOGNOSY IN TRADITIONAL AND MODERN SYSTEM OF MEDICINE.pptxNandakishor Bhaurao Deshmukh
 
Speech, hearing, noise, intelligibility.pptx
Speech, hearing, noise, intelligibility.pptxSpeech, hearing, noise, intelligibility.pptx
Speech, hearing, noise, intelligibility.pptxpriyankatabhane
 
Radiation physics in Dental Radiology...
Radiation physics in Dental Radiology...Radiation physics in Dental Radiology...
Radiation physics in Dental Radiology...navyadasi1992
 
Base editing, prime editing, Cas13 & RNA editing and organelle base editing
Base editing, prime editing, Cas13 & RNA editing and organelle base editingBase editing, prime editing, Cas13 & RNA editing and organelle base editing
Base editing, prime editing, Cas13 & RNA editing and organelle base editingNetHelix
 
ALL ABOUT MIXTURES IN GRADE 7 CLASS PPTX
ALL ABOUT MIXTURES IN GRADE 7 CLASS PPTXALL ABOUT MIXTURES IN GRADE 7 CLASS PPTX
ALL ABOUT MIXTURES IN GRADE 7 CLASS PPTXDole Philippines School
 
Harmful and Useful Microorganisms Presentation
Harmful and Useful Microorganisms PresentationHarmful and Useful Microorganisms Presentation
Harmful and Useful Microorganisms Presentationtahreemzahra82
 

Último (20)

Pests of safflower_Binomics_Identification_Dr.UPR.pdf
Pests of safflower_Binomics_Identification_Dr.UPR.pdfPests of safflower_Binomics_Identification_Dr.UPR.pdf
Pests of safflower_Binomics_Identification_Dr.UPR.pdf
 
User Guide: Capricorn FLX™ Weather Station
User Guide: Capricorn FLX™ Weather StationUser Guide: Capricorn FLX™ Weather Station
User Guide: Capricorn FLX™ Weather Station
 
Observational constraints on mergers creating magnetism in massive stars
Observational constraints on mergers creating magnetism in massive starsObservational constraints on mergers creating magnetism in massive stars
Observational constraints on mergers creating magnetism in massive stars
 
Thermodynamics ,types of system,formulae ,gibbs free energy .pptx
Thermodynamics ,types of system,formulae ,gibbs free energy .pptxThermodynamics ,types of system,formulae ,gibbs free energy .pptx
Thermodynamics ,types of system,formulae ,gibbs free energy .pptx
 
GENERAL PHYSICS 2 REFRACTION OF LIGHT SENIOR HIGH SCHOOL GENPHYS2.pptx
GENERAL PHYSICS 2 REFRACTION OF LIGHT SENIOR HIGH SCHOOL GENPHYS2.pptxGENERAL PHYSICS 2 REFRACTION OF LIGHT SENIOR HIGH SCHOOL GENPHYS2.pptx
GENERAL PHYSICS 2 REFRACTION OF LIGHT SENIOR HIGH SCHOOL GENPHYS2.pptx
 
FREE NURSING BUNDLE FOR NURSES.PDF by na
FREE NURSING BUNDLE FOR NURSES.PDF by naFREE NURSING BUNDLE FOR NURSES.PDF by na
FREE NURSING BUNDLE FOR NURSES.PDF by na
 
Call Girls in Majnu Ka Tilla Delhi 🔝9711014705🔝 Genuine
Call Girls in Majnu Ka Tilla Delhi 🔝9711014705🔝 GenuineCall Girls in Majnu Ka Tilla Delhi 🔝9711014705🔝 Genuine
Call Girls in Majnu Ka Tilla Delhi 🔝9711014705🔝 Genuine
 
Pests of Blackgram, greengram, cowpea_Dr.UPR.pdf
Pests of Blackgram, greengram, cowpea_Dr.UPR.pdfPests of Blackgram, greengram, cowpea_Dr.UPR.pdf
Pests of Blackgram, greengram, cowpea_Dr.UPR.pdf
 
Fertilization: Sperm and the egg—collectively called the gametes—fuse togethe...
Fertilization: Sperm and the egg—collectively called the gametes—fuse togethe...Fertilization: Sperm and the egg—collectively called the gametes—fuse togethe...
Fertilization: Sperm and the egg—collectively called the gametes—fuse togethe...
 
(9818099198) Call Girls In Noida Sector 14 (NOIDA ESCORTS)
(9818099198) Call Girls In Noida Sector 14 (NOIDA ESCORTS)(9818099198) Call Girls In Noida Sector 14 (NOIDA ESCORTS)
(9818099198) Call Girls In Noida Sector 14 (NOIDA ESCORTS)
 
basic entomology with insect anatomy and taxonomy
basic entomology with insect anatomy and taxonomybasic entomology with insect anatomy and taxonomy
basic entomology with insect anatomy and taxonomy
 
Microteaching on terms used in filtration .Pharmaceutical Engineering
Microteaching on terms used in filtration .Pharmaceutical EngineeringMicroteaching on terms used in filtration .Pharmaceutical Engineering
Microteaching on terms used in filtration .Pharmaceutical Engineering
 
Topic 9- General Principles of International Law.pptx
Topic 9- General Principles of International Law.pptxTopic 9- General Principles of International Law.pptx
Topic 9- General Principles of International Law.pptx
 
User Guide: Pulsar™ Weather Station (Columbia Weather Systems)
User Guide: Pulsar™ Weather Station (Columbia Weather Systems)User Guide: Pulsar™ Weather Station (Columbia Weather Systems)
User Guide: Pulsar™ Weather Station (Columbia Weather Systems)
 
THE ROLE OF PHARMACOGNOSY IN TRADITIONAL AND MODERN SYSTEM OF MEDICINE.pptx
THE ROLE OF PHARMACOGNOSY IN TRADITIONAL AND MODERN SYSTEM OF MEDICINE.pptxTHE ROLE OF PHARMACOGNOSY IN TRADITIONAL AND MODERN SYSTEM OF MEDICINE.pptx
THE ROLE OF PHARMACOGNOSY IN TRADITIONAL AND MODERN SYSTEM OF MEDICINE.pptx
 
Speech, hearing, noise, intelligibility.pptx
Speech, hearing, noise, intelligibility.pptxSpeech, hearing, noise, intelligibility.pptx
Speech, hearing, noise, intelligibility.pptx
 
Radiation physics in Dental Radiology...
Radiation physics in Dental Radiology...Radiation physics in Dental Radiology...
Radiation physics in Dental Radiology...
 
Base editing, prime editing, Cas13 & RNA editing and organelle base editing
Base editing, prime editing, Cas13 & RNA editing and organelle base editingBase editing, prime editing, Cas13 & RNA editing and organelle base editing
Base editing, prime editing, Cas13 & RNA editing and organelle base editing
 
ALL ABOUT MIXTURES IN GRADE 7 CLASS PPTX
ALL ABOUT MIXTURES IN GRADE 7 CLASS PPTXALL ABOUT MIXTURES IN GRADE 7 CLASS PPTX
ALL ABOUT MIXTURES IN GRADE 7 CLASS PPTX
 
Harmful and Useful Microorganisms Presentation
Harmful and Useful Microorganisms PresentationHarmful and Useful Microorganisms Presentation
Harmful and Useful Microorganisms Presentation
 

Activity Recognition from Accelerometers in Smart Homes

  • 1. Activity Recognition from Accelerometers in Smart Homes Tom Diethe Manager, Core Machine Learning Amazon Research Cambridge Honorary Research Fellow Intelligent Systems Laboratory University of Bristol Bristol Robotics Lab, 16th August 2017
  • 2. Introduction Methods Bayesian Dictionary Learning Infer.NET implementation Experiments Conclusions References About Me 1996-2000 B.Sc. Experimental Psychology @ UoB 2000-2005 QinetiQ Ltd 2005-2006 M.Sc. Intelligent Systems @ UCL 2006-2010 Ph.D. Machine Learning & Signal Processing 2010-2012 Post-doc x2 @ UCL (CS + Stats) 2012-2013 Consultant @ BMJ Group 2013-2014 Researcher @ Microsoft Research Cambridge 2014-2017 Research Fellow @ UoB (SPHERE) Next stop: Amazon Research Cambridge Tom Diethe Amazon Research Cambridge; University of Bristol Activity Recognition from Accelerometers in Smart Homes
  • 3. Introduction Methods Bayesian Dictionary Learning Infer.NET implementation Experiments Conclusions References The SPHERE project Environmental temp, light, humidity, air quality, water & electricity Video emotion, gait, activity, interaction Wearable activity, sleep, etc. Contextual information demographics, medical history diaries, annotations Tom Diethe Amazon Research Cambridge; University of Bristol Activity Recognition from Accelerometers in Smart Homes
  • 4. Introduction Methods Bayesian Dictionary Learning Infer.NET implementation Experiments Conclusions References The SPHERE project Environmental temp, light, humidity, air quality, water & electricity Video emotion, gait, activity, interaction Wearable activity, sleep, etc. Contextual information demographics, medical history diaries, annotations Tom Diethe Amazon Research Cambridge; University of Bristol Activity Recognition from Accelerometers in Smart Homes
  • 5. Introduction Methods Bayesian Dictionary Learning Infer.NET implementation Experiments Conclusions References Goal: Activity Recognition in Smart Homes Noisy data −→ methods that characterise the noise Tom Diethe Amazon Research Cambridge; University of Bristol Activity Recognition from Accelerometers in Smart Homes
  • 6. Introduction Methods Bayesian Dictionary Learning Infer.NET implementation Experiments Conclusions References Goal: Activity Recognition in Smart Homes Noisy data −→ methods that characterise the noise Missing packets −→ robust methods Tom Diethe Amazon Research Cambridge; University of Bristol Activity Recognition from Accelerometers in Smart Homes
  • 7. Introduction Methods Bayesian Dictionary Learning Infer.NET implementation Experiments Conclusions References Goal: Activity Recognition in Smart Homes Noisy data −→ methods that characterise the noise Missing packets −→ robust methods Limited storage and transmission capacity −→ sparse/compressive methods Tom Diethe Amazon Research Cambridge; University of Bristol Activity Recognition from Accelerometers in Smart Homes
  • 8. Introduction Methods Bayesian Dictionary Learning Infer.NET implementation Experiments Conclusions References Goal: Activity Recognition in Smart Homes Noisy data −→ methods that characterise the noise Missing packets −→ robust methods Limited storage and transmission capacity −→ sparse/compressive methods Labelled data time-consuming and costly to acquire Tom Diethe Amazon Research Cambridge; University of Bristol Activity Recognition from Accelerometers in Smart Homes
  • 9. Introduction Methods Bayesian Dictionary Learning Infer.NET implementation Experiments Conclusions References Goal: Activity Recognition in Smart Homes Noisy data −→ methods that characterise the noise Missing packets −→ robust methods Limited storage and transmission capacity −→ sparse/compressive methods Labelled data time-consuming and costly to acquire −→ Active Learning Tom Diethe Amazon Research Cambridge; University of Bristol Activity Recognition from Accelerometers in Smart Homes
  • 10. Introduction Methods Bayesian Dictionary Learning Infer.NET implementation Experiments Conclusions References Goal: Activity Recognition in Smart Homes Noisy data −→ methods that characterise the noise Missing packets −→ robust methods Limited storage and transmission capacity −→ sparse/compressive methods Labelled data time-consuming and costly to acquire −→ Active Learning Differing training and deployment settings Tom Diethe Amazon Research Cambridge; University of Bristol Activity Recognition from Accelerometers in Smart Homes
  • 11. Introduction Methods Bayesian Dictionary Learning Infer.NET implementation Experiments Conclusions References Goal: Activity Recognition in Smart Homes Noisy data −→ methods that characterise the noise Missing packets −→ robust methods Limited storage and transmission capacity −→ sparse/compressive methods Labelled data time-consuming and costly to acquire −→ Active Learning Differing training and deployment settings −→ Transfer Learning Tom Diethe Amazon Research Cambridge; University of Bristol Activity Recognition from Accelerometers in Smart Homes
  • 12. Introduction Methods Bayesian Dictionary Learning Infer.NET implementation Experiments Conclusions References Motivation Tom Diethe Amazon Research Cambridge; University of Bristol Activity Recognition from Accelerometers in Smart Homes
  • 13. Introduction Methods Bayesian Dictionary Learning Infer.NET implementation Experiments Conclusions References Motivation Would like a sparse representation of the data Tom Diethe Amazon Research Cambridge; University of Bristol Activity Recognition from Accelerometers in Smart Homes
  • 14. Introduction Methods Bayesian Dictionary Learning Infer.NET implementation Experiments Conclusions References Motivation Would like a sparse representation of the data Don’t want to make parametric assumptions Tom Diethe Amazon Research Cambridge; University of Bristol Activity Recognition from Accelerometers in Smart Homes
  • 15. Introduction Methods Bayesian Dictionary Learning Infer.NET implementation Experiments Conclusions References Motivation Would like a sparse representation of the data Don’t want to make parametric assumptions Features for classification (activity recognition) Tom Diethe Amazon Research Cambridge; University of Bristol Activity Recognition from Accelerometers in Smart Homes
  • 16. Introduction Methods Bayesian Dictionary Learning Infer.NET implementation Experiments Conclusions References Motivation Would like a sparse representation of the data Don’t want to make parametric assumptions Features for classification (activity recognition) −→ Dictionary Learning Tom Diethe Amazon Research Cambridge; University of Bristol Activity Recognition from Accelerometers in Smart Homes
  • 17. Introduction Methods Bayesian Dictionary Learning Infer.NET implementation Experiments Conclusions References Motivation Would like a sparse representation of the data Don’t want to make parametric assumptions Features for classification (activity recognition) −→ Dictionary Learning Would like to incorporate with other models Tom Diethe Amazon Research Cambridge; University of Bristol Activity Recognition from Accelerometers in Smart Homes
  • 18. Introduction Methods Bayesian Dictionary Learning Infer.NET implementation Experiments Conclusions References Motivation Would like a sparse representation of the data Don’t want to make parametric assumptions Features for classification (activity recognition) −→ Dictionary Learning Would like to incorporate with other models −→ Bayesian approach Tom Diethe Amazon Research Cambridge; University of Bristol Activity Recognition from Accelerometers in Smart Homes
  • 19. Introduction Methods Bayesian Dictionary Learning Infer.NET implementation Experiments Conclusions References Dictionary Learning Aim: find a set of vectors di , (dictionary), to represent x ∈ Rn as a linear combination of these vectors: x = k i=1 zi di s.t. k n. (1) Tom Diethe Amazon Research Cambridge; University of Bristol Activity Recognition from Accelerometers in Smart Homes
  • 20. Introduction Methods Bayesian Dictionary Learning Infer.NET implementation Experiments Conclusions References Dictionary Learning Aim: find a set of vectors di , (dictionary), to represent x ∈ Rn as a linear combination of these vectors: x = k i=1 zi di s.t. k n. (1) Sparsity: few non-zero components zi (or many close to 0) Tom Diethe Amazon Research Cambridge; University of Bristol Activity Recognition from Accelerometers in Smart Homes
  • 21. Introduction Methods Bayesian Dictionary Learning Infer.NET implementation Experiments Conclusions References Dictionary Learning Aim: find a set of vectors di , (dictionary), to represent x ∈ Rn as a linear combination of these vectors: x = k i=1 zi di s.t. k n. (1) Sparsity: few non-zero components zi (or many close to 0) Cost function for m input vectors arranged in columns of matrix X ∈ Rn×m min Z,D X − DZ 2 F + λ n i=1 Ω(zi ) s.t. di 2 ≤ C, ∀i = 1, . . . , k. where D ∈ Rn×k is the dictionary, Z ∈ Rk×m are the coefficients, and Ω(·) is a sparsity inducing regulariser, and λ determines the relative importance of good reconstructionsTom Diethe Amazon Research Cambridge; University of Bristol Activity Recognition from Accelerometers in Smart Homes
  • 22. Introduction Methods Bayesian Dictionary Learning Infer.NET implementation Experiments Conclusions References Generative model for Equation (1) x N dot d z 0 0 ⌧ a b N N Ga(1, 1) Ga n m k Based on Yang et al. (2015). Tom Diethe Amazon Research Cambridge; University of Bristol Activity Recognition from Accelerometers in Smart Homes
  • 23. Introduction Methods Bayesian Dictionary Learning Infer.NET implementation Experiments Conclusions References Generative model for Equation (1) x N dot d z 0 0 ⌧ a b N N Ga(1, 1) Ga Ga(1, 1) n m k Based on Yang et al. (2015). Gamma prior over β, which allows the dictionary atoms to be automatically scaled Tom Diethe Amazon Research Cambridge; University of Bristol Activity Recognition from Accelerometers in Smart Homes
  • 24. Introduction Methods Bayesian Dictionary Learning Infer.NET implementation Experiments Conclusions References Generative model for Equation (1) x N dot d z ↵ 0 ⌧ a b N N Ga(1, 1) Ga N(0, 1) Ga(1, 1) n m k Based on Yang et al. (2015). Gamma prior over β, which allows the dictionary atoms to be automatically scaled additional level of hierarchy through the variables α on the means of the dictionary components, which can aid online learning Tom Diethe Amazon Research Cambridge; University of Bristol Activity Recognition from Accelerometers in Smart Homes
  • 25. Introduction Methods Bayesian Dictionary Learning Infer.NET implementation Experiments Conclusions References Generative model for Equation (1) x N dot d z ↵ 0 ⌧ a b N N Ga(1, 1) Ga N(0, 1) Ga(1, 1) n m k Based on Yang et al. (2015). Gamma prior over β, which allows the dictionary atoms to be automatically scaled additional level of hierarchy through the variables α on the means of the dictionary components, which can aid online learning a, b control sparsity: Ga(1, 1) → non-sparse Ga(0.5, 10− 6) → sparse Tom Diethe Amazon Research Cambridge; University of Bristol Activity Recognition from Accelerometers in Smart Homes
  • 26. Introduction Methods Bayesian Dictionary Learning Infer.NET implementation Experiments Conclusions References Infer.NET Mode (.NET code) Infer.NET Compiler Inference code Answers!Data Framework for running Bayesian inference in graphical modelsa Experiments used Monob, running on OS-X and Linux Inference engine: Variational Message Passing (VMP), deterministic approximation algorithm a http://research.microsoft.com/infernet b http://www.mono-project.com/ Tom Diethe Amazon Research Cambridge; University of Bristol Activity Recognition from Accelerometers in Smart Homes
  • 27. Introduction Methods Bayesian Dictionary Learning Infer.NET implementation Experiments Conclusions References Modelling code p u b l i c void BDL( double [ , ] x , i n t numBases , double a , double b ) { var b a s i s = new Range ( numBases ) ; var s i g n a l = new Range ( x . GetLength ( 0 ) ) ; var sample = new Range ( x . GetLength ( 1 ) ) ; var n o i s e P r e c i s i o n = V a r i a b l e . GammaFromShapeAndRate (1 , 1 ) ; var c o e f f i c i e n t P r e c i s i o n s = V a r i a b l e . Array<double >(s i g n a l , b a s i s ) ; var dictionaryMeans = V a r i a b l e . Array<double >(b a s i s , sample ) ; var d i c t i o n a r y P r e c i s i o n s = V a r i a b l e . Array<double >(b a s i s , sample ) ; var c o e f f i c i e n t s = V a r i a b l e . Array<double >(s i g n a l , b a s i s ) ; var d i c t i o n a r y = V a r i a b l e . Array<double >(b a s i s , sample ) ; var s i g n a l s = V a r i a b l e . Array<double >(s i g n a l , sample ) ; dictionaryMeans [ b a s i s ] [ sample ] = V a r i a b l e . GaussianFromMeanAndVariance (0 , 1 ) . ForEach ( b a d i c t i o n a r y P r e c i s i o n s [ b a s i s , sample ] = V a r i a b l e . GammaFromShapeAndRate (1 , 1 ) . ForEach ( b a s c o e f f i c i e n t P r e c i s i o n s [ s i g n a l , b a s i s ] = V a r i a b l e . GammaFromShapeAndRate ( a , b ) . ForEach ( s i c o e f f i c i e n t s [ s i g n a l , b a s i s ] = V a r i a b l e . GaussianFromMeanAndPrecision (0 , c o e f f i c i e n t P r e c d i c t i o n a r y [ b a s i s , sample ] = V a r i a b l e . GaussianFromMeanAndPrecision ( dictionaryMeans [ b a s i s , sample ] , d i c t i o n a r y P r e c i s i o n s [ b a s i s , sample ] ) ; d i c t i o n a r y [ b a s i s , sample ] . I n i t i a l i s e T o ( d i c t i o n a r y P r i o r s [ b a s i s , sample ] ) ; var c l e a n S i g n a l s = V a r i a b l e . M a t r i x M u l t i p l y ( c o e f f i c i e n t s , d i c t i o n a r y ) . Named( ” c l e a n ” ) ; s i g n a l s [ s i g n a l , sample ] = V a r i a b l e . GaussianFromMeanAndPrecision ( c l e a n S i g n a l s [ s i g n a l , s s i g n a l s . ObservedValue = x ; } Tom Diethe Amazon Research Cambridge; University of Bristol Activity Recognition from Accelerometers in Smart Homes
  • 28. Introduction Methods Bayesian Dictionary Learning Infer.NET implementation Experiments Conclusions References Datasets HAR dataset Anguita et al. (2013) Publicly available activity recognition dataset1 Smart-phone on the waist, with tri-axial accelerometer, 50 Hz Annotation was done using video-recordings. Activities: Lie, Stand, Walk, Sit, Ascend stairs, Descend stairs SPHERE challenge dataset Twomey et al. (2016) collected by SPHERE project and made public as a challenge2 Tri-axial accelerometer on dominant wrist, 20 Hz, range ±8 g Activites: Lie, Stand, Walk 1 https://archive.ics.uci.edu/ml/datasets/Human+Activity+ Recognition+Using+Smartphones 2 http://irc-sphere.ac.uk/sphere-challenge/home Tom Diethe Amazon Research Cambridge; University of Bristol Activity Recognition from Accelerometers in Smart Homes
  • 29. Introduction Methods Bayesian Dictionary Learning Infer.NET implementation Experiments Conclusions References Sparsity Non-sparse priors (Ga(1, 1)) average sparsity 0.84 bases signals Sparse priors (Ga(0.5, 10−6)) average sparsity 0.96 bases signals Tom Diethe Amazon Research Cambridge; University of Bristol Activity Recognition from Accelerometers in Smart Homes
  • 30. Introduction Methods Bayesian Dictionary Learning Infer.NET implementation Experiments Conclusions References Convergence of VMP Model evidence converges monotonically to a local maximum 0 2 4 6 8 10 12 14 16 18 Iterations 350000 300000 250000 200000 150000 100000 50000 0 50000 Evidencelogodds non-sparse sparse Tom Diethe Amazon Research Cambridge; University of Bristol Activity Recognition from Accelerometers in Smart Homes
  • 31. Introduction Methods Bayesian Dictionary Learning Infer.NET implementation Experiments Conclusions References Convergence of reconstruction error 0 2 4 6 8 10 12 14 16 18 Iterations 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 RMSE non-sparse sparse Tom Diethe Amazon Research Cambridge; University of Bristol Activity Recognition from Accelerometers in Smart Homes
  • 32. Introduction Methods Bayesian Dictionary Learning Infer.NET implementation Experiments Conclusions References Learnt Dictionary 150000 100000 50000 0 50000 100000 150000 150000 100000 50000 0 50000 100000 150000 150000 100000 50000 0 50000 100000 150000 20020406080100120140160 150000 100000 50000 0 50000 100000 150000 20020406080100120140160 20020406080100120140160 20020406080100120140160 Dictionary 0-16 x y Figure: 16 example bases from the dictionary of 128 bases inferred by BDL using the non-sparse priors on HAR dataset Tom Diethe Amazon Research Cambridge; University of Bristol Activity Recognition from Accelerometers in Smart Homes
  • 33. Introduction Methods Bayesian Dictionary Learning Infer.NET implementation Experiments Conclusions References Coefficients Tom Diethe Amazon Research Cambridge; University of Bristol Activity Recognition from Accelerometers in Smart Homes
  • 34. Introduction Methods Bayesian Dictionary Learning Infer.NET implementation Experiments Conclusions References Reconstructions 0.6 0.4 0.2 0.0 0.2 0.4 0.6 0.8 0.0 0.5 1.0 1.5 2.0 2.5 3.0 0.6 0.4 0.2 0.0 0.2 0.4 0.6 0.8 signal reconstruction ±s.d. Reconstructions, RMSE=0.0276 time (s) magnitude Tom Diethe Amazon Research Cambridge; University of Bristol Activity Recognition from Accelerometers in Smart Homes
  • 35. Introduction Methods Bayesian Dictionary Learning Infer.NET implementation Experiments Conclusions References Reconstruction Performance SPAMS BDL Sparse BDL Bases RMSE Sparsity RMSE Sparsity RMSE Sparsity 64 0.0480 0.71 0.0519 0.88 0.0293 0.62 128 0.0457 0.84 0.0400 0.94 0.0276 0.84 256 0.0438 0.92 0.0316 0.99 0.0288 0.93 512 0.0423 0.96 0.0224 0.99 0.0231 0.96 Tom Diethe Amazon Research Cambridge; University of Bristol Activity Recognition from Accelerometers in Smart Homes
  • 36. Introduction Methods Bayesian Dictionary Learning Infer.NET implementation Experiments Conclusions References Classification Multi-class Bayes Point Machine (BPM) Herbrich et al. (2001), also using Infer.NET Assumptions: Feature values x are always fully observed The order of instances does not matter The predictive distribution is a linear discriminant of the form: p(yi |xi , w) = p(yi |si = w xi ), where w are the weights and si is the score for instance i y ˜s s 1 N w x ⇥ µw ⌧w N N D A R Tom Diethe Amazon Research Cambridge; University of Bristol Activity Recognition from Accelerometers in Smart Homes
  • 37. Introduction Methods Bayesian Dictionary Learning Infer.NET implementation Experiments Conclusions References Activity Recognition Results Use coefficients as features of classification algorithm 64 bases + bias feature Metric: per-class 1 vs rest area under ROC curve HAR dataset Activity BDL sparse BDL SPAMS Walking 0.73 0.83 0.88 Ascending stairs 0.63 0.60 0.83 Descending stairs 0.61 0.34 0.82 Sitting 0.74 0.72 0.89 Standing 0.51 0.43 0.98 Lying down 0.95 0.95 0.95 Average 0.70 0.65 0.89 SPHERE dataset Activity BDL sparse BDL SPAMS Walking 0.55 0.51 0.46 Standing 0.69 0.62 0.44 Lying down 0.86 0.85 0.46 Average 0.70 0.66 0.45 Tom Diethe Amazon Research Cambridge; University of Bristol Activity Recognition from Accelerometers in Smart Homes
  • 38. Introduction Methods Bayesian Dictionary Learning Infer.NET implementation Experiments Conclusions References Transfer Learning Source: DS = {(xi , yi )mS i=1} Target: DT = {(xi , yi )mT i=1} Assumptions: labels available for source domain labels can be acquired for target domain, but costly Tom Diethe Amazon Research Cambridge; University of Bristol Activity Recognition from Accelerometers in Smart Homes
  • 39. Introduction Methods Bayesian Dictionary Learning Infer.NET implementation Experiments Conclusions References Hierarchical Model Multiclass Bayes Point Machine (BPM) y ˜s s 1 N w x ⇥ µw ⌧w N N D A R Tom Diethe Amazon Research Cambridge; University of Bristol Activity Recognition from Accelerometers in Smart Homes
  • 40. Introduction Methods Bayesian Dictionary Learning Infer.NET implementation Experiments Conclusions References Hierarchical Model Community: Learn community posteriors y ˜s s 1 N w x ⇥ µw ⌧w µµ µ k⌧ ✓⌧ N N N D A R Tom Diethe Amazon Research Cambridge; University of Bristol Activity Recognition from Accelerometers in Smart Homes
  • 41. Introduction Methods Bayesian Dictionary Learning Infer.NET implementation Experiments Conclusions References Hierarchical Model Transfer: weight means and precisions replaced by community posteriors from source y ˜s s 1 N w x ⇥ µw ⌧w N N D A R Tom Diethe Amazon Research Cambridge; University of Bristol Activity Recognition from Accelerometers in Smart Homes
  • 42. Introduction Methods Bayesian Dictionary Learning Infer.NET implementation Experiments Conclusions References Hierarchical Model Personalisation: smoothly evolve from generic to custom predictions y ˜s s 1 N w x ⇥ µw ⌧w N N D A R Tom Diethe Amazon Research Cambridge; University of Bristol Activity Recognition from Accelerometers in Smart Homes
  • 43. Introduction Methods Bayesian Dictionary Learning Infer.NET implementation Experiments Conclusions References Active Learning Baseline: random (= online learning) Uncertainty Sampling Nearest to decision boundary, fails in case of corrupted data Certainty Sampling Furthest from decision boundary. Normally nonsensical but solves corruption issue Value of Information (VOI) Kapoor et al. (2007) Decision theoretic measure that uses p(y = 1) vs p(y = 0) (relative risk). Expensive to compute. Tom Diethe Amazon Research Cambridge; University of Bristol Activity Recognition from Accelerometers in Smart Homes
  • 44. Introduction Methods Bayesian Dictionary Learning Infer.NET implementation Experiments Conclusions References Datasets Source: Anguita et al. (2013) 30 subjects, Smartphone, 50Hz, Video annotations Target: Zhang and Sawchuk (2012) 14 subjects, MotionNode, 100Hz, Observer annotations Classes: Walking upstairs vs. Walking downstairs Tom Diethe Amazon Research Cambridge; University of Bristol Activity Recognition from Accelerometers in Smart Homes
  • 45. Introduction Methods Bayesian Dictionary Learning Infer.NET implementation Experiments Conclusions References Results 0 5 10 15 20 # instances 0.0 0.2 0.4 0.6 0.8 1.0Accuracy Hold-out Accuracy Random US CS VOI+ VOI- Tom Diethe Amazon Research Cambridge; University of Bristol Activity Recognition from Accelerometers in Smart Homes
  • 46. Introduction Methods Bayesian Dictionary Learning Infer.NET implementation Experiments Conclusions References Conclusions and Further Work Improved methods for Bayesian Dictionary Learning Sparsity does not need to be enforced Application to accelerometer signals for Activity Recognition Hierarchical extension to BPM for transfer learning Combination of Active & Transfer learning −→ fast learning on target dataset Uncertainty Sampling good enough Tom Diethe Amazon Research Cambridge; University of Bristol Activity Recognition from Accelerometers in Smart Homes
  • 47. Introduction Methods Bayesian Dictionary Learning Infer.NET implementation Experiments Conclusions References Conclusions and Further Work Improved methods for Bayesian Dictionary Learning Sparsity does not need to be enforced Application to accelerometer signals for Activity Recognition Hierarchical extension to BPM for transfer learning Combination of Active & Transfer learning −→ fast learning on target dataset Uncertainty Sampling good enough Further Work: Incorporate classifier directly into the model Tom Diethe Amazon Research Cambridge; University of Bristol Activity Recognition from Accelerometers in Smart Homes
  • 48. Introduction Methods Bayesian Dictionary Learning Infer.NET implementation Experiments Conclusions References Conclusions and Further Work Improved methods for Bayesian Dictionary Learning Sparsity does not need to be enforced Application to accelerometer signals for Activity Recognition Hierarchical extension to BPM for transfer learning Combination of Active & Transfer learning −→ fast learning on target dataset Uncertainty Sampling good enough Further Work: Incorporate classifier directly into the model Source code: https://github.com/IRC-SPHERE/bayesian-dictionary-learning https://github.com/IRC-SPHERE/ActiveTransfer Tom Diethe Amazon Research Cambridge; University of Bristol Activity Recognition from Accelerometers in Smart Homes
  • 49. Introduction Methods Bayesian Dictionary Learning Infer.NET implementation Experiments Conclusions References Selected References D Anguita, A Ghio, L Oneto, X Parra, and JL Reyes-Ortiz. A public domain dataset for human activity recognition using smartphones. In ESANN, 2013. Tom Diethe, Niall Twomey, and Peter Flach. Active transfer learning for activity recognition. In 24th European Symposium on Artificial Neural Networks, ESANN 2016, Bruges, Belgium, April 27-29, 2016, 2016a. Tom Diethe, Niall Twomey, and Peter Flach. BDL. NET: Bayesian dictionary learning in Infer. NET. In Machine Learning for Signal Processing (MLSP), 2016 IEEE 26th International Workshop on, pages 1–6. IEEE, 2016b. Ralf Herbrich, Thore Graepel, and Colin Campbell. Bayes point machines. Journal of Machine Learning Research, 1:245–279, January 2001. URL http://research.microsoft.com/apps/pubs/default.aspx?id=65611. Ashish Kapoor, Eric Horvitz, and Sumit Basu. Selective supervision: Guiding supervised learning with decision-theoretic active learning. In IJCAI, volume 7, pages 877–882, 2007. Niall Twomey, Tom Diethe, Meelis Kull, Hao Song, Massimo Camplani, Sion Hannuna, Xenofon Fafoutis, Ni Zhu, Pete Woznowski, Peter Flach, and Ian Craddock. The sphere challenge: Activity recognition with multimodal sensor data. arXiv preprint arXiv:1603.00797, 2016. Linxiao Yang, Jun Fang, Hong Cheng, and Hongbin Li. Sparse Bayesian dictionary learning with a Gaussian hierarchical model. CoRR, abs/1503.02144, 2015. URL http://arxiv.org/abs/1503.02144. M. Zhang and A.A. Sawchuk. USC-HAD: A daily activity dataset for ubiquitous activity recognition using wearable sensors. In ACM Int. Conf. on Ubiquitous Computing Workshop on Situation, Activity and Goal Awareness (SAGAware), 2012. Tom Diethe Amazon Research Cambridge; University of Bristol Activity Recognition from Accelerometers in Smart Homes