๐ข Columbia University
Towards a 'Universal Translator' for Neural Dynamics at Single-Cell, Single-Spike Resolution
ยท2778 wordsยท14 minsยท
loading
ยท
loading
Machine Learning
Self-Supervised Learning
๐ข 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,โฆ
The Fine-Grained Complexity of Gradient Computation for Training Large Language Models
ยท336 wordsยท2 minsยท
loading
ยท
loading
Natural Language Processing
Large Language Models
๐ข Columbia University
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.
The Fairness-Quality Tradeoff in Clustering
ยท2122 wordsยท10 minsยท
loading
ยท
loading
AI Generated
AI Theory
Fairness
๐ข Columbia University
Novel algorithms trace the optimal balance between clustering quality and fairness, revealing all non-dominated solutions for various objectives.
SemCoder: Training Code Language Models with Comprehensive Semantics Reasoning
ยท1972 wordsยท10 minsยท
loading
ยท
loading
Natural Language Processing
Large Language Models
๐ข Columbia University
SEMCODER: A novel 6.7B parameter code LLM surpasses GPT-3.5-turboโs performance on code generation and execution reasoning by employing โmonologue reasoningโโtraining the model to verbally explain codโฆ
Randomized Strategic Facility Location with Predictions
ยท1312 wordsยท7 minsยท
loading
ยท
loading
AI Theory
Optimization
๐ข Columbia University
Randomized strategies improve truthful learning-augmented mechanisms for strategic facility location, achieving better approximations than deterministic methods.
Promoting Fairness Among Dynamic Agents in Online-Matching Markets under Known Stationary Arrival Distributions
ยท1572 wordsยท8 minsยท
loading
ยท
loading
AI Generated
AI Theory
Fairness
๐ข Columbia University
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-leโฆ
Partial Transportability for Domain Generalization
ยท2485 wordsยท12 minsยท
loading
ยท
loading
AI Theory
Generalization
๐ข Columbia University
This paper introduces a novel technique to bound prediction risks in new domains using causal diagrams, enabling reliable AI performance guarantees.
Nonparametric Instrumental Variable Regression through Stochastic Approximate Gradients
ยท1348 wordsยท7 minsยท
loading
ยท
loading
AI Theory
Causality
๐ข Columbia University
SAGD-IV: a novel functional stochastic gradient descent algorithm for stable nonparametric instrumental variable regression, excelling in handling binary outcomes and various loss functions.
Mind the Gap: A Causal Perspective on Bias Amplification in Prediction & Decision-Making
ยท1773 wordsยท9 minsยท
loading
ยท
loading
AI Theory
Fairness
๐ข Columbia University
AI bias amplification in decision-making is uncovered, showing how fair prediction scores can become discriminatory after thresholding, urging stronger regulatory oversight.
Local Anti-Concentration Class: Logarithmic Regret for Greedy Linear Contextual Bandit
ยท2759 wordsยท13 minsยท
loading
ยท
loading
Machine Learning
Reinforcement Learning
๐ข Columbia University
Greedy algorithms for linear contextual bandits achieve poly-logarithmic regret under the novel Local Anti-Concentration condition, expanding applicable distributions beyond Gaussians and uniforms.
Is Cross-validation the Gold Standard to Estimate Out-of-sample Model Performance?
ยท1790 wordsยท9 minsยท
loading
ยท
loading
AI Theory
Optimization
๐ข Columbia University
Cross-validation isnโt always superior; simple plug-in methods often perform equally well for estimating out-of-sample model performance, especially when considering computational costs.
Inductive biases of multi-task learning and finetuning: multiple regimes of feature reuse
ยท3248 wordsยท16 minsยท
loading
ยท
loading
AI Generated
Machine Learning
Transfer Learning
๐ข Columbia University
Multi-task learning and finetuning show surprising feature reuse biases, including a novel โnested feature selectionโ regime where finetuning prioritizes a sparse subset of pretrained features, signifโฆ
Group-wise oracle-efficient algorithms for online multi-group learning
ยท316 wordsยท2 minsยท
loading
ยท
loading
AI Theory
Fairness
๐ข Columbia University
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 Secretaries with Unfair Predictions
ยท1586 wordsยท8 minsยท
loading
ยท
loading
AI Theory
Fairness
๐ข Columbia University
Fair algorithms can leverage biased predictions to improve performance while guaranteeing fairness for all candidates.
Extensive-Form Game Solving via Blackwell Approachability on Treeplexes
ยท2500 wordsยท12 minsยท
loading
ยท
loading
Reinforcement Learning
๐ข Columbia University
First algorithmic framework for Blackwell approachability on treeplexes, enabling stepsize-invariant EFG solvers with state-of-the-art convergence rates.
Disentangling Interpretable Factors with Supervised Independent Subspace Principal Component Analysis
ยท3550 wordsยท17 minsยท
loading
ยท
loading
AI Generated
Machine Learning
Representation Learning
๐ข Columbia University
Supervised Independent Subspace PCA (sisPCA) disentangles interpretable factors in high-dimensional data by leveraging supervision to maximize subspace dependence on target variables while minimizing โฆ
Disentangled Representation Learning in Non-Markovian Causal Systems
ยท2882 wordsยท14 minsยท
loading
ยท
loading
AI Theory
Causality
๐ข Columbia University
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: Model Selection And Linear-Time Inference
ยท1804 wordsยท9 minsยท
loading
ยท
loading
Machine Learning
Gaussian Processes
๐ข Columbia University
Computation-Aware Gaussian Processes (CaGP) achieve linear-time inference and model selection, enabling efficient training of GPs on large datasets without compromising uncertainty quantification, surโฆ
Community Detection Guarantees using Embeddings Learned by Node2Vec
ยท2609 wordsยท13 minsยท
loading
ยท
loading
AI Generated
AI Theory
Representation Learning
๐ข Columbia University
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.
Causal Imitation for Markov Decision Processes: a Partial Identification Approach
ยท1601 wordsยท8 minsยท
loading
ยท
loading
Machine Learning
Reinforcement Learning
๐ข Columbia University
This paper presents novel causal imitation learning algorithms using partial identification to achieve expert performance even when unobserved confounders affect Markov Decision Processes.