ABCDEFGHIJKLMNOPQRSTUV
1
YanProject Topic Project Description (tentative)Reading ListTentative Concrete Deadlineswhat students will get from this projectMember #1 Please sign upMember #2 Please sign up
2
3
1Space-efficient compound operations for Tensor Networks and Tensorial Neural Networkshttps://www.cs.umd.edu/class/fall2018/cmsc498V/slides/TensorBasics.pdf
https://arxiv.org/abs/1805.10352
http://www.ece.lsu.edu/jxr/papers-pdf/pldi02.pdf
https://arxiv.org/abs/1606.05696		Tahseen RabbaniAlex Reustle
4
2
5
3Adversarial ExamplePing-yeh Chiang
6
4PAC-RL (tabular)- Near-optimal regret bounds for reinforcement learning: http://www.jmlr.org/papers/volume11/jaksch10a/jaksch10a.pdf
- Sample complexity of episodic fixed-horizon reinforcement learning https://papers.nips.cc/paper/5827-sample-complexity-of-episodic-fixed-horizon-reinforcement-learning.pdf
- Minimax regret bounds for reinforcement learning https://arxiv.org/pdf/1703.05449.pdf
- Unifying PAC and Regret: Uniform PAC Bounds for Episodic Reinforcement Learning http://papers.nips.cc/paper/7154-unifying-pac-and-regret-uniform-pac-bounds-for-episodic-reinforcement-learning.pdf
- Policy Certificates: Towards Accountable Reinforcement Learning https://arxiv.org/pdf/1811.03056.pdf
- Optimistic posterior sampling for reinforcement learning: worst-case regret bounds https://papers.nips.cc/paper/6718-optimistic-posterior-sampling-for-reinforcement-learning-worst-case-regret-bounds.pdf
- Regret Bounds for Reinforcement Learning via Markov Chain Concentration https://arxiv.org/pdf/1808.01813.pdf
- Is Q-learning Provably Efficient? https://arxiv.org/pdf/1807.03765.pdf		Peng WanYanchao Sun
7
5Differentially Private Topic Model
8
6Spectral methods for exploration in POMDPs- Reinforcement Learning of POMDPs using Spectral Methods http://proceedings.mlr.press/v49/azizzadenesheli16a.pdf
- A PAC RL Algorithm for Episodic POMDPs https://arxiv.org/pdf/1605.08062.pdf		Alex LevineMarina Knittel
9
7Generalization of Neural NetworksSweta AgrawalSamyadeep
10
8Unsupervised BoostingLillian HuangJamie Matthews
11
9Off-policy evaluation- Eligibility traces for off-policy policy evaluation https://scholarworks.umass.edu/cgi/viewcontent.cgi?article=1079&context=cs_faculty_pubs
- Doubly Robust Off-policy Value Evaluation for Reinforcement Learning http://nanjiang.cs.illinois.edu/files/ICML2016-DR.pdf
- Optimal and Adaptive Off-Policy Evaluation in Contextual Bandits http://proceedings.mlr.press/v70/wang17a/wang17a.pdf
- Data-efficient off-policy policy evaluation for reinforcement learning http://proceedings.mlr.press/v48/thomasa16.pdf
- Using Options and Covariance Testing for Long Horizon Off-Policy Policy Evaluation http://papers.nips.cc/paper/6843-using-options-and-covariance-testing-for-long-horizon-off-policy-policy-evaluation.pdf
- Effective Evaluation Using Logged Bandit Feedback from Multiple Loggers https://arxiv.org/pdf/1703.06180.pdf
- Balanced Policy Evaluation and Learning http://papers.nips.cc/paper/8105-balanced-policy-evaluation-and-learning.pdf
- Breaking the Curse of Horizon: Infinite-Horizon Off-Policy Estimation http://papers.nips.cc/paper/7781-breaking-the-curse-of-horizon-infinite-horizon-off-policy-estimation.pdf		Zeyad EmamDaniel Lichy
12
10POMDPs and predictive state representations(PSRs)- Point-based value iteration: An anytime algorithm for POMDPs http://www.fore.robot.cc/papers/Pineau03a.pdf
- Perseus: Randomized Point-based Value Iteration for POMDPs
- Closing the learning–planning loop with predictive state representations
- Efficient Learning and Planning with Compressed Predictive States https://arxiv.org/abs/1312.0286
- Hilbert Space Embeddings of Hidden Markov Models https://www.researchgate.net/profile/Alexander_Smola/publication/221345134_Hilbert_Space_Embeddings_of_Hidden_Markov_Models/links/0912f5105f839ae5a9000000/Hilbert-Space-Embeddings-of-Hidden-Markov-Models.pdf
- Low-Rank Spectral Learning with Weighted Loss Functions http://nanjiang.cs.illinois.edu/files/lowrank_aistats15.pdf
- Completing State Representations using Spectral Learning http://nanjiang.cs.illinois.edu/files/nips18-psrf.pdf		Goonwanth
13
11Automatic searching of neural network structure
14
12Defense against Adversarial ExamplesUse a provable differentiable lower bound to the closest adversarial example for defense against adversarial exampleshttps://arxiv.org/abs/1902.01235Sahil Singla
15
13Bounding Fairness transfer between tasksGoal: Transfer of Machine Learning Fairness across Tasks. In the context of fairness, we are interested in the setting of a single domain which may have multiple tasks.This might exist in the context of transfer learning or in multi-task learning (MTL).1. A Survey of Transfer Learning, Weiss et al. [2016] 2. A Survey of Transfer Learning, Pan and Yang [2010] 3. A Survey on Multi-Task Learning, Zhang and Yang [2017] 4. A Model of Inductive Bias Learning, Baxter [2000] 5. A notion of task relatedness yielding provable multiple-task learning guarantees, Ben-David and Borbely [2008] 6. Algorithm-Dependent Generalization Bounds for Multi-Task Learning, Liu et al. [2017] : 7. A Model of Inductive Bias Learning, Baxter [2000] 8. A framework for learning predictive structures from multiple tasks and unlabeled data, Ando and 9. Bounds for linear multi-task learning, Maurer [2006a]: 10. The Rademacher complexity of linear transformation classes, Maurer [2006b] 11. The Benefit of Multitask Representation Learning, Maurer et al. [2016] 12. Estimating relatedness via data compression, Juba [2006] 13. Equality of opportunity in supervised learning, Hardt et al. [2016]Samuel Dooley
16
14Interpreting Neural Networks using similarity between the layersUse Canonical Correlation Analysis (used to compute similarity between NN layers) to emperically assess its potential as an alternative to validation set for tuning hyperparameters and architectural settings, investigate the scope of its use in structural search, as a possible metric to explain the 'overparameterization' phenomenon, and model compression. Post empericial investigations, go for theoretical underpinnings of CCA and why it is able to explain learning dynamics or our results.- SVCCA: Singular Vector Canonical Correlation
Analysis for Deep Learning Dynamics and
Interpretability
https://arxiv.org/pdf/1706.05806.pdf
- Insights on representational similarity in neural
networks with canonical correlation
http://papers.nips.cc/paper/7815-insights-on-representational-similarity-in-neural-networks-with-canonical-correlation.pdf
- Understanding Learning Dynamics Of Language Models with SVCCA
https://arxiv.org/pdf/1811.00225.pdf
- A CLOSER LOOK AT DEEP LEARNING HEURISTICS:
LEARNING RATE RESTARTS, WARMUP AND DISTILLATION
https://arxiv.org/pdf/1810.13243.pdf
- Neural Ordinary Differential Equations
https://arxiv.org/pdf/1806.07366.pdf
- Towards Understanding the Role of Over-Parametrization
in Generalization of Neural Networks
https://arxiv.org/pdf/1805.12076.pdf
- https://arxiv.org/pdf/1805.10352.pdf		Pranav GoelAnshul Shah
17
15Online Boosting Leonidas Tsepenekas Eleftheria Briakou
18
16Adversarial Example-https://arxiv.org/pdf/1801.02610.pdf
the first paper is about generating an adversarial example using GAN
-https://arxiv.org/pdf/1804.02485.pdf
the second paper hinges on the first paper; it tries to generate a fortified neural network that is strong against the adversarial example.
-https://arxiv.org/pdf/1812.06570.pdf
this is another paper, which uses VAE to defense from adversarial neural network
-https://arxiv.org/pdf/1903.00585.pdf
similar idea. Using VAE. This one is called PUVAE (purifying VAE). These are easy defenses.
-https://arxiv.org/pdf/1806.04646.pdf
this paper provides how to attack VAE		Nao RhoFan Yang
19
17Domain adaptationhttp://openaccess.thecvf.com/content_cvpr_2018/papers/Sankaranarayanan_Generate_to_Adapt_CVPR_2018_paper.pdfPrithviraj DharShlok Mishra
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100