1. ICML Learning Spatio-Temporal Structure from RGB-D Videos
for Human Activity Detection and Anticipation (@sla) Preferred
Infrastructure
2. ICML2013 l l T-PRIMAL20-30 l PFI l l NIPS l PFI 2
3. l ICML2013 l l RGB-D Agenda
4. ICML2013: 4
5. ICML2013: 5
6. ICML2013: 1. Machine Learning at Scale with GraphLab by
Carlos Guestrin l GAS(Gather-Apply-Scatter) l GraphLab2GraphLab3 2.
High-dimensional Sampling Algorithms and their Applications by
Santosh Vempala l Convex, Convex, and Convex 3. Acoustic Modeling
and Deep Learning for Speech Recognition by Vincent Vanhoucke
(Google Voice Search) l Deep Learning l Deep Belief Networks
[Bengio+, 2007] l GPGPU l Data(+dropout) l 1011HintonGoogle 6
7. ICML2013Sparse, Deep, and Random 7 Sparse, Random,
MultiBandit KernelSVMReinforcementBayesian
http://www.machinedlearnings.com/2013/06/icml-2013-sparse-deep-and-random.html
8. l @sla : "Learning Spatio-Temporal Structure from RGB-D
Videos for Human Activity... l @beam2d: "Local Deep Kernel Learning
for Efficient Non-linear SVM Prediction l @conditional: "Vanishing
Component Analysis l @jkomiyama_ : "Active Learning for
Multi-Objective Optimization l @kisa12012 : "Large-Scale Learning
with Less RAM via Randomization l @Quasi_quant2010 : "Topic
Discovery through Data Dependent and Random Projections l @tabe2314
: "Fast Image Tagging l @unnonouno : "ELLA: An Efficient Lifelong
Learning Algorithm l @sleepy_yoshi : "Distributed Training of
Large-scale Logistic Models" 8 ---Sparse ---Deep (?) ---Random
---Others(Spatio-Temporal, Component Analysis,
Multi-taskDistributed)
9. l ICML2013 l l RGB-D Agenda
10. i.i.d l l l i.i.d l i.i.d l 10
11. : l 3 l Robot Learning l Machine Learning with Test-Time
Budgets l Learning with Sequential Models l l Cost-sensitive
Learning l Imitation Learning / Interactive Learning l
Reinforcement Learning l Imperative Learning (Data
Search/Aggregation) l l Anytime l l l Feature11
12. : DAgger [Ross+, AISTATS11] l Dataset Aggregator 12 Ross et
al., A Reduction of Imitation Learning and Structured Prediction to
No-Regret Online Learning, AISTATS'11
13. l ICML2013 l l RGB-D Agenda
14. Learning Spatio-Temporal Structure from RGB-D Videos for
Human Activity Detection and Anticipation" l 1: Hemi Koppula
(Cornell University) l 2: Ashtosh Saxena (Cornell University) l
Andrew NgRobot/CV l Robot LearningInvited Talk l RGB-D l l Activity
Detection: l Sub-activity l Activity Anticipation: 14
15. l http://pr.cs.cornell.edu/anticipation/ 15
16. l Cutting-plane training of structural SVMs [Joachims+,
MLJ2009] 16