Paper presentations
Each lecture, we will review two papers from the reading list. Each paper will be presented by a student who will also be responsible to lead the in-class discussion. When it is your turn to present, I will ask you to upload your slides to Canvas (in PDF format, one slide per page) by 8am of the lecture day (20% grade penalty for late submission). You should aim for a 40 minute presentation. A rule of thumb is to devote about 20 minutes to describe the objectives of the work and the proposed technical solution. About 5-10 minutes should be dedicated to presenting the experimental results. Finally, you should prepare a 10 minute discussion highlighting the contributions of the work (what differentiates this paper from previous work?), its strengths as well as its weaknesses (technical, applicative, or experimental). Don't be afraid to be controversial or to ask questions/opinions to your classmates: it is your responsibility to lead an interactive discussion session and you are free to choose the style. Try to conclude your presentation with a list of suggested extensions of the work presented. Don't simply report the future work items discussed in the conclusion section of the paper: think independently about how you would choose to continue the research addressed in the article.
Written critiques
All students must upload a short written critique (about half a page of text) of each paper presented in class by 8am of the lecture day. Please summarize in a couple of sentences the objectives and the technical approach. Discuss in more detail the contributions, strengths and weaknesses of the work, exactly as if you were to present the paper in class. Don't forget to include your list of proposed extensions to the method. I will review all critiques every few weeks and provide detailed feedback. Each student can opt not to submit critiques for up to 3 papers without any penalty.
Reading list
Image Categorization
-
Deep Residual Learning for Image Recognition.
K. He, X. Zhang, S. Ren, and J. Sun.
CVPR, 2016. Best paper award.
Winner of the following challenges: ILSVRC 2015 classification, ILSVRC 2015 detection, ILSVRC 2015 localization, MS COCO 2015 detection.
-
Aggregated Residual Transformations for Deep Neural Networks.
S. Xie, R. Girshick, P. Dollar, Z. Tu, and K. He.
To apper at CVPR, 2017.
-
Spatial Transformer Networks.
M. Jaderberg, K. Simonyan, A. Zisserman, K. Kavukcuoglu.
NIPS, 2015.
Object Detection
-
SSD: Single Shot MultiBox Detector.
W. Liu, D. Anguelov, D. Erhan, C. Szegedy, S. Reed, C.-Y. Fu, A.C. Berg.
ECCV, 2016.
Semantic Segmentation
-
Fully Convolutional Instance-aware Semantic Segmentation.
Y. Li, H. Qi, J. Dai, X. Ji, and Y. Wei.
CVPR, 2017.
Winner of the COCO 2016 segmentation challenge.
Image compression
-
End-to-end Optimized Image Compression.
J. Balle, V. Laparra, and E.P. Simoncelli.
To appear at ICLR, 2017.
Video Analysis
-
UntrimmedNets for Weakly Supervised Action Recognition and Detection.
L. Wang, Y. Xiong, D. Lin, L. Van Gool.
CVPR 2017.
Generative Adversarial Networks
-
Generative Adversarial Nets.
I.J. Goodfellow, J. Pouget-Abadie, M. Mirza, B. Xu, D. Warde-Farley, S. Ozair, A. Courville, and Y. Bengio.
NIPS 2014.
-
Improved Techniques for Training GANs.
T. Salimans, I. Goodfellow, W. Zaremba, V. Cheung, A. Radford, and X. Chen.
NIPS, 2016.
Natural Language Processing
-
Show and Tell: A Neural Image Caption Generator.
O. Vinyals, A. Toshev, S. Bengio, D. Erhan.
CVPR, 2015.
-
Ask Me Anything: Dynamic Memory Networks for Natural Language Processing.
A. Kumar, O. Irsoy, P. Ondruska, M. Iyyer, J. Bradbury, I. Gulrajani, V. Zhong, R. Paulus, R. Socher.
ICML, 2016.
-
Learning End-to-End Goal-Oriented Dialog.
A. Bordes, Y-L. Boureau, J. Weston.
To appear at ICLR, 2017.
Reinforcement Learning
-
Value Iteration Networks.
A. Tamar, Y. Wu, G. Thomas, S. Levine, P. Abbeel.
NIPS, 2016.
Best paper award.
-
Reinforcement Learning with Unsupervised Auxiliary Tasks.
M. Jaderberg, V. Mnih, W.M. Czarnecki, T. Schaul, J.Z. Leibo, D. Silver, K. Kavukcuoglu.
To appear at ICLR, 2017.
-
Learning to Act by Predicting the Future.
A. Dosovitskiy, V, Koltun.
To appear at ICLR, 2017.
Learning to learn
-
Learning to learn by gradient descent by gradient descent.
M. Andrychowicz, M. Denil, S. Gomez, M.W. Hoffman, D. Pfau, T. Schaul, N. de Freitas.
NIPS, 2016.
-
Optimization as a Model for Few-Shot Learning.
S. Ravi, H. Larochelle.
To appear at ICLR, 2017.
-
Neural Architecture Search with Reinforcement Learning.
B. Zoph, Q. Le.
To appear at ICLR, 2017.
Model capacity, regularization, generalization
-
Understanding deep learning requires rethinking generalization.
C. Zhang, S. Bengio, M. Hardt, B. Recht, O. Vinyals.
To appear at ICLR, 2017.
Best paper award.
Sequence modeling
-
Using Fast Weights to Attend to the Recent Past.
J. Ba, G.E. Hinton, V. Mnih, J.Z. Leibo, C. Ionescu.
NIPS, 2016.