You are on page 1of 4

Course Information

Course Name Deep Learning Seminar


Semester Winter 2016/17
Location & Time University of Haifa
Education Building, Class 463, Tuesdays 10:15-11:45.
Lecturer Dr. Amir Adler
E-mail adleram@cs.technion.ac.il
Syllabus (1) Review of Deep Learning applications and architectures:
fully connected networks, convolutional networks (CNN),
recurrent networks (RNN), Stochastic Gradient Descent,
Back-Propagation, and Regularization Techniques.
(2) Students Lectures (40 minutes each), based on state-of-the-
art papers.
Pre-requisites: Machine Learning or Algorithms
Grading: 85% Lecture + 15% Class Presence & Participation

1
Course Plan

Topic Week

Introduction to Deep Learning: (1) Review of computer vison, natural language 1


processing, and speech processing applications. (2) Evolution of neural networks
technology, and milestones during past 10 years,
Machine Learning concepts review: supervised, unsupervised, semi-supervised 2
learning, representation learning, reinforcement learning.
Fully connected and convolutional networks in detail. Multi-layer perceptron, 3
non-linear activation units, SoftMax layer, convolutional layer, max-pooling,
average-pooling.
Stochastic Gradient Descent (SGD) and loss functions (mean-squared error, 4
negative log likelihood). SGD with momentum.
The Back-Propagation algorithm 5
Recurrent Neural Networks (RNN): Basic RNN structure, characters prediction 6
and image captioning with RNNs, exploding and vanishing gradients problem,
Long Short Term Memory (LSTM), Gated Recurrent Unit (GRU).
Students Lectures 7
Students Lectures 8
Students Lectures 9
Students Lectures 10
Students Lectures 11
Students Lectures 12
Students Lectures 13
Students Lectures 14

Text Book

1. Ian Goodfellow and Yoshua Bengio and Aaron Courville, "Deep Learning", MIT Press
2016. Available online: http://www.deeplearningbook.org/

2
Lectures

No. Paper Name Conference Link


/Journal
1 Beyond short snippets deep CVPR 2015 https://arxiv.org/abs/1503.08909
networks for video classification
large scale video classification with CVPR 2014 http://vision.stanford.edu/pdf/karpat
convolutional neural networks hy14.pdf

2 Deep Residual Learning for Image CVPR 2016 https://arxiv.org/abs/1512.03385


Recognition
Identity Mappings in Deep ECCV 2016 https://arxiv.org/abs/1603.05027
Residual Networks
3 SSD: Single Shot MultiBox ECCV 2016 https://arxiv.org/abs/1512.02325
Detector
Faster R-CNN: Towards Real-Time NIPS 2015 https://arxiv.org/abs/1506.01497
Object Detection with Region
Proposal Networks
4 Rich feature hierarchies for CVPR 2014 https://arxiv.org/abs/1311.2524
accurate object detection and
semantic segmentation
OverFeat: Integrated Recognition, ICLR 2014 https://arxiv.org/pdf/1312.6229.pdf
Localization and Detection using
Convolutional Networks
5 Show, Attend and Tell: Neural ICML 2014 https://arxiv.org/pdf/1502.03044v3.
Image Caption Generation with pdf
Visual Attention
Unifying Visual-Semantic NIPS 2014 https://arxiv.org/abs/1411.2539
Embeddings with Multimodal
Neural Language Models
6 Mastering the game of Go with Nature 2016 http://www.nature.com/nature/journ
deep neural networks and tree al/v529/n7587/full/nature16961.ht
search (by Google DeepMind) ml

Human level control through deep Nature 2015 http://www.nature.com/nature/journ


reinforcement learning (by Google al/v518/n7540/full/nature14236.ht
DeepMind) ml

7 Learning Hierarchical Features for PAMI 2013 http://yann.lecun.com/exdb/publis/p


Scene Labeling df/farabet-pami-13.pdf
Indoor Semantic Segmentation ICLR 2013 https://arxiv.org/abs/1301.3572
using depth information
8 Understanding the difficulty of AISTATS http://jmlr.org/proceedings/papers/v
training deep feedforward neural 2010 9/glorot10a/glorot10a.pdf
3
networks
Batch Normalization: Accelerating ICML 2015 http://jmlr.org/proceedings/papers/v
Deep Network Training by 37/ioffe15.pdf
Reducing
Internal Covariate Shift
9 Learning spatiotemporal features ICCV 2015 https://arxiv.org/pdf/1412.0767.pdf
with 3d convolutional networks
sequence to sequence video to text CVPR 2015 https://arxiv.org/abs/1505.00487

10 Very Deep Convolutional Networks ICLR 2015 https://arxiv.org/pdf/1409.1556.pdf


for Large-Scale Image Rcognition
Going Deeper With Convolutions CVPR 2015 https://arxiv.org/abs/1409.4842

11 Object Contour Detection with a CVPR 2016 http://www.cv-


Fully Convolutional Encoder- foundation.org/openaccess/content_
Decoder Network cvpr_2016/papers/Yang_Object_Co
ntour_Detection_CVPR_2016_pap
er.pdf

ECCV 2014 http://citeseerx.ist.psu.edu/viewdoc/


Learning a deep convolutional download?doi=10.1.1.642.1999&re
network for image super-resolution p=rep1&type=pdf

12 Explain Images with Multimodal NIPS 2014 https://arxiv.org/pdf/1410.1090.pdf


Recurrent Neural Networks Workshop
Long-term Recurrent Convolutional CVPR 2015 http://www.icsi.berkeley.edu/pubs/
Networks for Visual Recognition vision/longtermrecurrent15.pdf
and Description
13 Zero-Shot Learning Through Cross- NIPS 2013 http://nlp.stanford.edu/pubs/Socher
Modal Transfer GanjooManningNg_NIPS2013.pdf

DeViSE: A Deep Visual-Semantic NIPS 2013 http://static.googleusercontent.com/


Embedding Model media/research.google.com/en//pub
s/archive/41473.pdf

14 Video Super-Resolution With IEEE Trans. http://ivpl.eecs.northwestern.edu/sit


Convolutional Neural Networks on Comp. es/default/files/07444187.pdf
Imaging
2016
ICLR 2016 https://arxiv.org/pdf/1511.05440v6.
Deep multi-scale video prediction pdf
beyond mean square error

NIPS 2015 https://arxiv.org/pdf/1506.05439v3.


15 Learning with a Wasserstein Loss
pdf

ICCV 2009 http://www.ariel.ac.il/sites/ofirpele/


Fast and Robust Earth Mover’s publications/ICCV2009.pdf
Distances (background paper)

You might also like