AI Theory

Causal Dependence Plots (CDPs) visualize how machine learning model predictions causally depend on input features, overcoming limitations of existing methods that ignore causal relationships.

Pre-trained transformer models can predict human smell perception by encoding odorant chemical structures, aligning with expert labels, continuous ratings, and similarity assessments.

Neural operators’ continuous discretization is proven impossible in general Hilbert spaces, but achievable using strongly monotone operators, opening new avenues for numerical methods in scientific ma…

New efficient attack reveals GNN model training data properties.

This paper introduces a novel quantitative definition of AI alignment for social decision-making, proposing probably approximately aligned policies and a method to safeguard any autonomous agent’s act…

Boost classifier stability with the novel inflated argmax, guaranteeing reliable multiclass classification without distributional assumptions!

New model-agnostic framework for out-of-distribution generalization uses multicalibration across overlapping groups, showing improved robustness and prediction under various distribution shifts.

This paper delivers novel, universally applicable bounds for the smallest NTK eigenvalue, regardless of data distribution or dimension, leveraging the hemisphere transform.

BDNE: a novel boundary decomposition method accurately estimates the nadir objective vector in complex multi-objective optimization problems.

Bisimulation metrics and optimal transport distances are equivalent and can be computed efficiently using a novel Sinkhorn Value Iteration algorithm.

A novel model reveals how hippocampal-entorhinal circuits use compositional coding and modular attractor networks to enable robust and flexible spatial representation, advancing our understanding of c…

This paper presents a novel algorithm for binary search using distributional predictions, achieving optimal query complexity O(H(p) + log n) and demonstrating enhanced robustness against prediction er…

AI systems acquire bias during training, impacting accuracy across sub-populations. This research unveils bias’s dynamic nature, revealing how classifier preferences shift over time, influenced by dat…

This paper presents efficient algorithms to detect whether an agent updates beliefs optimally (Bayesian) or exhibits bias towards their prior beliefs, using information design and signaling schemes.

This paper presents a novel, UCB-like algorithm for bandits with stochastic and adversarial constraints, achieving optimal performance without the stringent assumptions of prior primal-dual methods.

This paper presents a novel method to make black box neural networks intervenable using only a small validation set with concept labels, improving the effectiveness of concept-based interventions.

Leaky ReLU networks exhibit benign overfitting under surprisingly relaxed conditions: input dimension only needs to linearly scale with sample size, challenging prior assumptions in the field.

Learners can improve accuracy in strategic classification by selectively revealing partial classifier information to agents, strategically guiding agent behavior and maximizing accuracy.

GraphComBO efficiently optimizes functions defined on node subsets within graphs using Bayesian Optimization. It tackles challenges posed by combinatorial complexity and computationally expensive fun…

Boost machine learning model robustness by minimizing a novel data-driven risk criterion that blends Bayesian nonparametrics and smooth ambiguity aversion, ensuring superior out-of-sample performance.