| 9:00 am - 9:15 am | Welcome to BayLearn 2020, BayLearn Organizers: Jerremy Holland,Jean-François Paiement, Sudarshan Lamkhede, Alice Xiang |
|
| 9:15 am - 10:00 am | Keynote 1: Timnit Gebru | |
| 10:00 am - 10:15 am | Q&A | |
| 10:15 am - 10:30 am | BREAK | |
| 10:30 am - 11:00 am | Keynote 2: Sandrine Dudoit | |
| 11:00 am - 11:15 am | Q&A | |
11:15 am - 11:55 pm
| Keynote 3: Chelsea Finn | |
| 11:55 am - 12:10 pm | Q&A | |
| 12:10 pm - 1:00 pm | LUNCH BREAK | |
| 1:00 pm - 1:30 pm | Keynote 4: Susan Athey | |
| 1:30 pm - 1:45 pm | Q&A | |
| 1:45 pm - 2:00 pm | BREAK | |
|
| |
| 2:00 pm - 3:00 pm | Poster Session I | |
ROOM # 1
| Cluster 1: Fairness, Explainable ML, Privacy, and Robustness | |
| Neural Additive Models: Interpretable Machine Learning with Neural Nets | |
| siVAE: interpreting latent dimensions within variational autoencoders | |
| Selectivity considered harmful: evaluating the causal impact of class selectivity in DNNs | |
| Synthetic Health Data for Fostering Reproducibility of Private Research Studies | |
| Adversarial Learning for Debiasing Knowledge Base Embeddings | Video Not Available
|
| Robustness Analysis of Deep Learning via Implicit Models | |
ROOM # 2
| Cluster 2: Computer Vision | |
| Protecting Against Image Translation Deepfakes by Leaking Universal Perturbations from Black-Box Neural Networks | |
| Anatomy of Catastrophic Forgetting: HiddenRepresentations and Task Semantics | |
| CoCon: Cooperative-Contrastive Learning | |
| Can Neural Networks Learn Non-Verbal Reasoning? | |
| Modality-Agnostic Attention Fusion for visual search with text feedback | |
ROOM # 3
| Cluster 3: Deep Learning | |
| Revisiting Spatial Invariance with Low-Rank Local Connectivity | |
| What is being transferred in transfer learning? | |
| Neural Anisotropy Directions | |
| Learning Discrete Energy-based Models via Auxiliary-variable Local Exploration | |
| What Neural Networks Memorize and Why: Discovering the Long Tail via Influence Estimation |
|
ROOM # 4
| Cluster 4: ML Methods and Tools | |
| Bandit-based Monte Carlo Optimization for Nearest Neighbors | |
| LassoNet: A Neural Network with Feature Sparsity | |
| Temperature check: theory and practice for training models with softmax-cross-entropy losses | |
| Meta-Learning Requires Meta-Augmentation | |
| Energy-based View of Retrosynthesis | |
ROOM # 5
| Cluster 5: Reinforcement Learning | |
| Learning to grow: control of materials self-assembly using evolutionary reinforcement learning | |
| Deployment-Efficient Reinforcement Learning via Model-Based Offline Optimization | |
| Provably Efficient Policy Optimization via Thompson Sampling | |
| Uncovering Task Clusters in Multi-Task Reinforcement Learning | |
| Curriculum and Decentralized Learning in Google Research Football | |
| 3:00 pm - 4:00 pm | Poster Session II | |
ROOM # 5
| Cluster 6: Bayesian Learning and Uncertainty | |
| Autofocused oracles for design | |
| Exact posteriors of wide Bayesian neural networks | |
| Deep Ensembles: a loss landscape perspective | |
| Active Online Domain Adaptation | |
| TSGLR: an Adaptive Thompson Sampling for the Switching Multi-Armed Bandit Problem | |
| ROOM # 2 | Cluster 7: Computer Vision and Robotics | |
| Interpretable Planning-Aware Representations for Multi-Agent Trajectory Forecasting | |
| Learning Mixed-Integer Convex Optimization Strategies for Robot Planning and Control | |
| Beyond Supervision for Monocular Depth Estimation | |
| A Synthetic Data Petri Dish for Studying Mode Collapse in GANs | |
| Attention-Sampling Graph Convolutional Networks | |
| Towards Learning Robots Which Adapt On The Fly | |
ROOM # 4
| Cluster 8: Deep ML and other topics | |
| Simultaneous Learning of the Inputs and Parameters in Neural Collaborative Filtering | |
| Ads Clickthrough Rate Prediction Models For Multi-Datasource Tasks | |
| Neural Interventional GRU-ODEs | |
ROOM # 3
| Cluster 9: Optimization | |
| VP-FO: A Variable Projection Method for Training Neural Networks | |
| Optimizing Memory Placement using Evolutionary Graph Reinforcement Learning | |
| Neural Representations in Hybrid Recommender Systems: Prediction versus Regularization | |
| ECLIPSE: An Extreme-Scale Linear Program Solver for Web-Applications | |
| ROOM # 1 | Cluster 10: Reinforcement Learning | |
| Safety Aware Reinforcement Learning (SARL) | |
| Meta Attention Networks: Meta Learning Attention to Modulate Information Between Sparsely Interacting Recurrent Modules | |
| Batch Reinforcement Learning Through Continuation Method | |
| See, Hear, Explore: Curiosity via Audio-Visual Association | |
| 4:00 pm - 5:00 pm | Poster Session III | |
ROOM # 4
| Cluster 11: Natural Language Processing | |
| Automated Utterance Generation | |
| Entity Skeletons as Intermediate Representations for Visual Storytelling | |
| Learning to reason by learning on rationales | |
| MUFASA: Multimodal Fusion Architecture Search for Electronic Health Records | |
| VirAAL: Virtual Adversarial Active Learning | |
| ChemBERTa: Utilizing Transformer-Based Attention for Understanding Chemistry | |
ROOM # 5
| Cluster 12: On-Device ML and Human-Computer Interaction | |
| GANs for Continuous Path Keyboard Input Modeling | |
| Architecture Compression | |
| A flexible, extensible software framework for model compression based on the LC algorithm | |
| Rotation-Invariant Gait Identification with Quaternion Convolutional Neural Networks | |
| ROOM # 2 | Cluster 13: Large-Scale Learning | |
| Heteroskedastic and Imbalanced Deep Learning with Adaptive Regularization | |
| Self-supervised Learning for Deep Models in Recommendations | |
| Learning Multi-granular Quantized Embeddings for Large-Vocab Categorical Features in Recommender Systems | |
| Distributed Sketching Methods for Privacy Preserving Regression | |
| Hamming Space Locality Preserving Neural Hashing for Similarity Search | |
ROOM # 3
| Cluster 14: Optimization | |
| Exact Polynomial-time Convex Optimization Formulations for Two-Layer ReLU Networks | |
| DisARM: An Antithetic Gradient Estimator for Binary Latent Variables | |
| Boosted Sparse Oblique Decision Trees | |
| Whitening and second order optimization both destroy information about the dataset, and can make generalization impossible | |
