🏢 Columbia University

A new self-supervised learning approach, Multi-task Masking (MtM), significantly improves the prediction accuracy of neural population activity by capturing neural dynamics at multiple spatial scales,…

New research precisely defines the computational limits of training large language models, revealing a sharp threshold based on parameter matrix entries, paving the way for faster algorithms.

Novel algorithms trace the optimal balance between clustering quality and fairness, revealing all non-dominated solutions for various objectives.

SEMCODER: A novel 6.7B parameter code LLM surpasses GPT-3.5-turbo’s performance on code generation and execution reasoning by employing ‘monologue reasoning’—training the model to verbally explain cod…

Randomized strategies improve truthful learning-augmented mechanisms for strategic facility location, achieving better approximations than deterministic methods.

This paper presents novel algorithms for online matching markets that prioritize fairness among dynamic agents, achieving asymptotic optimality in various scenarios and offering extensions to group-le…

This paper introduces a novel technique to bound prediction risks in new domains using causal diagrams, enabling reliable AI performance guarantees.

SAGD-IV: a novel functional stochastic gradient descent algorithm for stable nonparametric instrumental variable regression, excelling in handling binary outcomes and various loss functions.

AI bias amplification in decision-making is uncovered, showing how fair prediction scores can become discriminatory after thresholding, urging stronger regulatory oversight.

Greedy algorithms for linear contextual bandits achieve poly-logarithmic regret under the novel Local Anti-Concentration condition, expanding applicable distributions beyond Gaussians and uniforms.

Cross-validation isn’t always superior; simple plug-in methods often perform equally well for estimating out-of-sample model performance, especially when considering computational costs.

Multi-task learning and finetuning show surprising feature reuse biases, including a novel ’nested feature selection’ regime where finetuning prioritizes a sparse subset of pretrained features, signif…

Oracle-efficient algorithms conquer online multi-group learning, achieving sublinear regret even with massive, overlapping groups, paving the way for fair and efficient large-scale online systems.

Fair algorithms can leverage biased predictions to improve performance while guaranteeing fairness for all candidates.

First algorithmic framework for Blackwell approachability on treeplexes, enabling stepsize-invariant EFG solvers with state-of-the-art convergence rates.

Supervised Independent Subspace PCA (sisPCA) disentangles interpretable factors in high-dimensional data by leveraging supervision to maximize subspace dependence on target variables while minimizing …

This paper introduces graphical criteria and an algorithm for disentangling causal factors from heterogeneous data in non-Markovian settings, advancing causal representation learning.

Computation-Aware Gaussian Processes (CaGP) achieve linear-time inference and model selection, enabling efficient training of GPs on large datasets without compromising uncertainty quantification, sur…

Node2Vec, a popular network embedding method, is proven to consistently recover community structure in stochastic block models, paving the way for more reliable unsupervised community detection.

This paper presents novel causal imitation learning algorithms using partial identification to achieve expert performance even when unobserved confounders affect Markov Decision Processes.