Workshop
Physical Reasoning and Inductive Biases for the Real World
Krishna Murthy Jatavallabhula · Rika Antonova · Kevin Smith · Hsiao-Yu Tung · Florian Shkurti · Jeannette Bohg · Josh Tenenbaum
Tue 14 Dec, 8 a.m. PST
Much progress has been made on end-to-end learning for physical understanding and reasoning. If successful, understanding and reasoning about the physical world promises far-reaching applications in robotics, machine vision, and the physical sciences. Despite this recent progress, our best artificial systems pale in comparison to the flexibility and generalization of human physical reasoning.
Neural information processing systems have shown promising empirical results on synthetic datasets, yet do not transfer well when deployed in novel scenarios (including the physical world). If physical understanding and reasoning techniques are to play a broader role in the physical world, they must be able to function across a wide variety of scenarios, including ones that might lie outside the training distribution. How can we design systems that satisfy these criteria?
Our workshop aims to investigate this broad question by bringing together experts from machine learning, the physical sciences, cognitive and developmental psychology, and robotics to investigate how these techniques may one day be employed in the real world. In particular, we aim to investigate the following questions: 1. What forms of inductive biases best enable the development of physical understanding techniques that are applicable to real-world problems? 2. How do we ensure that the outputs of a physical reasoning module are reasonable and physically plausible? 3. Is interpretability a necessity for physical understanding and reasoning techniques to be suitable to real-world problems?
Unlike end-to-end neural architectures that distribute bias across a large set of parameters, modern structured physical reasoning modules (differentiable physics, relational learning, probabilistic programming) maintain modularity and physical interpretability. We will discuss how these inductive biases might aid in generalization and interpretability, and how these techniques impact real-world problems.
Schedule
Tue 8:00 a.m. - 8:15 a.m.
|
Introductory remarks
(
Live talk
)
>
SlidesLive Video |
🔗 |
Tue 8:15 a.m. - 8:45 a.m.
|
Tomer Ullman
(
Live talk
)
>
SlidesLive Video |
🔗 |
Tue 8:45 a.m. - 9:15 a.m.
|
Nils Thuerey
(
Live talk
)
>
SlidesLive Video |
Nils Thuerey 🔗 |
Tue 9:15 a.m. - 9:45 a.m.
|
Karen Liu
(
Live talk
)
>
SlidesLive Video |
Karen Liu 🔗 |
Tue 10:30 a.m. - 10:40 a.m.
|
Playful Interactions for Representation Learning
(
Oral
)
>
SlidesLive Video |
Sarah Young · Pieter Abbeel · Lerrel Pinto 🔗 |
Tue 10:40 a.m. - 10:50 a.m.
|
Efficient and Interpretable Robot Manipulation with Graph Neural Networks
(
Oral
)
>
SlidesLive Video |
Yixin Lin · Austin Wang · Eric Undersander · Akshara Rai 🔗 |
Tue 10:50 a.m. - 11:00 a.m.
|
Vision-based system identification and 3D keypoint discovery using dynamics constraints
(
Oral
)
>
SlidesLive Video |
Miguel Jaques · Martin Asenov · Michael Burke · Timothy Hospedales 🔗 |
Tue 11:00 a.m. - 11:02 a.m.
|
3D Neural Scene Representations for Visuomotor Control
(
Spotlight
)
>
SlidesLive Video |
Yunzhu Li · Shuang Li · Vincent Sitzmann · Pulkit Agrawal · Antonio Torralba 🔗 |
Tue 11:02 a.m. - 11:04 a.m.
|
Learning Graph Search Heuristics
(
Spotlight
)
>
SlidesLive Video |
Michal Pándy · Rex Ying · Gabriele Corso · Petar Veličković · Jure Leskovec · Pietro Liò 🔗 |
Tue 11:04 a.m. - 11:06 a.m.
|
Efficient Partial Simulation Quantitatively Explains Deviations from Optimal Physical Predictions
(
Spotlight
)
>
SlidesLive Video |
Ilona Bass · Kevin Smith · Elizabeth Bonawitz · Tomer Ullman 🔗 |
Tue 11:06 a.m. - 11:08 a.m.
|
TorchDyn: Implicit Models and Neural Numerical Methods in PyTorch
(
Spotlight
)
>
SlidesLive Video |
Michael Poli · Stefano Massaroli · Atsushi Yamashita · Hajime Asama · Jinkyoo Park · Stefano Ermon 🔗 |
Tue 11:08 a.m. - 11:10 a.m.
|
3D-OES: Viewpoint-Invariant Object-FactorizedEnvironment Simulators
(
Spotlight
)
>
SlidesLive Video |
Hsiao-Yu Tung · Zhou Xian · Mihir Prabhudesai · Katerina Fragkiadaki 🔗 |
Tue 11:10 a.m. - 11:12 a.m.
|
DLO@Scale: A Large-scale Meta Dataset for Learning Non-rigid Object Pushing Dynamics
(
Spotlight
)
>
SlidesLive Video |
Robert Gieselmann · Danica Kragic · Florian T. Pokorny · Alberta Longhini 🔗 |
Tue 11:12 a.m. - 11:14 a.m.
|
AVoE: A Synthetic 3D Dataset on Understanding Violation of Expectation for Artificial Cognition
(
Spotlight
)
>
SlidesLive Video |
Arijit Dasgupta · Jiafei Duan · Marcelo Ang Jr · Cheston Tan 🔗 |
Tue 11:14 a.m. - 11:16 a.m.
|
Physics-guided Learning-based Adaptive Control on the SE(3) Manifold
(
Spotlight
)
>
SlidesLive Video |
Thai Duong · Nikolay Atanasov 🔗 |
Tue 11:16 a.m. - 11:18 a.m.
|
Neural NID Rules
(
Spotlight
)
>
SlidesLive Video |
Luca Viano · Johanni Brea 🔗 |
Tue 11:30 a.m. - 12:00 p.m.
|
Kelsey Allen
(
Live talk
)
>
SlidesLive Video |
Kelsey Allen 🔗 |
Tue 12:00 p.m. - 12:30 p.m.
|
Kyle Cranmer
(
Live talk
)
>
SlidesLive Video |
Kyle Cranmer 🔗 |
Tue 12:30 p.m. - 1:00 p.m.
|
Shuran Song
(
Live talk
)
>
SlidesLive Video |
Shuran Song 🔗 |
Tue 1:00 p.m. - 2:00 p.m.
|
Industry Panel: Kenneth Tran (Koidra), Hiro Ono (NASA JPL), Aleksandra Faust (Google Brain), Michael Roberts (COVID-19 AIX-COVNET University of Cambridge)
(
Discussion Panel
)
>
SlidesLive Video |
🔗 |
Tue 2:00 p.m. - 2:45 p.m.
|
Research Panel
(
Discussion Panel
)
>
SlidesLive Video |
🔗 |
Tue 2:45 p.m. - 4:00 p.m.
|
Social - GatherTown ( GatherTown Meeting ) > link | 🔗 |