Interpretable Machine Learning in Natural and Social Sciences (2022)
2022-06-27..2022-06-29
Berkeley, United States
who it's for
machine learning researchers, computational neuroscientists, computational biologists, computational chemists and biophysicists, physical science data scientists, platform engineers supporting audits, legal scholars and policy researchers
topics
causal feature learning; human-AI team objectives; Rashomon set model selection; sensitivity analyses for under-specification; Girsanov reweighting importance sampling; Brownian bridge trajectory sampling; symmetrized atom density descriptors; spherical harmonics atomic descriptors; dilated convolutional models for genomics; DeepLIFT and FastISM attribution; banded ridge regression for voxels; platform-supported social media auditing
| title | author | length |
|---|---|---|
| Goals And Interpretable Variables In Neuroscience | David Danks | 46m |
Argues that which variables become interpretable depends on the goal: four concrete construction methods — clustering by intercorrelation (parcellations)… | ||
| Accuracy and Interpretability Through the Lens of Human-AI Teaming | Sina Fazelpour | 31m |
Argues that treating models as components of human-AI teams flips the accuracy versus interpretability trade-off: the most predictive model can harm system… | ||
| Panel on Societal Dimensions of Explanation | David Danks, Sina Fazelpour | 01h 06m |
Panel argues that interpretability must be tied to concrete aims and context, not lofty redefinitions: some defend a framework with contextual free parameters… | ||
| Conceptual Challenges In Connecting Interpretability And Causality | Alex D'Amour | 39m |
Argues that connecting interpretability to causality requires explicit contextual assumptions and that prediction alone is insufficient. Demonstrates… | ||
| Explanation: A(N Abridged) Survey | Joe Halpern | 39m |
Proposes a formal definition of (causal) explanation: a fact the agent does not already know for certain that, if true, would be an actual cause of the… | ||
| Panel on Causality | Alex D'Amour, Joe Halpern | 49m |
Panelists argue causality lives in models: claims are about what an interventionist causality model predicts, which interventions are allowed, and which… | ||
| Interpreting Machine Learning From the Perspective of Nonequilibrium Systems | David Limmer | 31m |
Presents a concrete method that turns rare-event sampling into an importance-sampling problem over trajectories by learning an optimal drift using policy… | ||
| Interpretability In Atomic-Scale Machine Learning | Michele Ceriotti | 31m |
Claims that atom-scale ML uses representations built from symmetrized correlations of the atom density and a hierarchy of 2-, 3‑body (distance+angle)… | ||
| Panel on Interpretability in the Physical Sciences | Michele Ceriotti, David Limmer | 57m |
Panel shows concrete methods: variational optimization of an optimal force (solution of a backward/Governorov-type equation) to drive and analyze rare… | ||
| Interpreting Deep Learning Models Of Functional Genomics Data To Decode Regulatory Sequence... | Anshul Kundaje | 38m |
Dilated convolutional neural networks map DNA sequence to genome-wide molecular profiles at single-base resolution across hundreds to thousands of cell… | ||
| Machine Learning, Deep Networks and Interpretability in Systems, Cognitive and... | Jack Gallant | 45m |
Argues the mammalian brain behaves like a dynamical deep network but neuroscience is severely data-limited, so two concrete methods are used: extract… | ||
| Panel on Interpretability in the Biological Sciences | Debbie Marks, Anshul Kundaje, Jack Gallant | 43m |
Genome-scale work treats a three billion base input where true positive binding sites number 2,000–20,000 positive sites per factor, yet neural nets trained on… | ||
| Interpretability and Algorithmic Fairness | Alice Xiang | 34m |
Two short projects: one formalizes causal fairness with a potential outcomes framework that shifts interventions from immutable attributes to the perception of… | ||
| Legal Barriers To Interpretable Machine Learning | Rebecca Wexler | 34m |
Machine learning scrutiny in criminal cases runs into four concrete legal barriers: trade secret claims that let developers withhold source code, training… | ||
| Platform-supported Auditing Of Social Media Algorithms For Public Interest | Aleksandra Korolova | 32m |
Researchers become an advertiser and run paired campaigns (e.g. lumber vs cleaning, Domino’s vs Instacart, Netflix vs NVIDIA) to control for usage, cost and… | ||
| Panel on Interpretability in the Law | Alice Xiang, Aleksandra Korolova, Rebecca Wexler | 01h 06m |
Panel documents specific failures and partial fixes: Facebook ad delivery shows discriminatory outcomes while LinkedIn results rely on a workaround for missing… | ||
| nothing matches | ||