New metrics KSS and KPS are introduced to evaluate multilingual machine unlearning quality and cross-language consistency in LLMs, addressing limitations of single-language evaluation protocols.
the Journal of machine Learning research , volume=
11 Pith papers cite this work. Polarity classification is still indexing.
11
Pith papers citing it
citation-role summary
method 1
citation-polarity summary
roles
method 1polarities
use method 1representative citing papers
A new auditing framework measures how much of the public's submitted viewpoints is lost in AI-generated consultation summaries, finding that official summaries represent the population worse than a random set of participants and that critical voices are most likely excluded.
Cosmological MBHBs coalesce in ~1 Gyr with high eccentricities; scaling relations from 30 Griffin re-simulations link dynamical-friction, hardening and total times to galaxy and orbital properties.
SGD is reformulated via a master equation from discrete updates, producing a discrete Fokker-Planck equation that predicts non-stationary variance growth proportional to learning rate in flat Hessian directions.
A generalized variance-reduced ZO hard-thresholding algorithm removes prior limits on random directions for gradient estimates, yielding improved convergence rates under standard assumptions.
DKPS-based methods predict new model benchmark scores using cached responses, matching baseline mean absolute error with substantially fewer queries and an offline query selection approach.
Any fixed integer linear program with a finite feasible set can be answered by a precomputed linear decision tree using polynomially many arithmetic operations per cost query; a practical construction works on small instances.
Empirical analysis shows scaling inference compute via strategies like tree search can be more efficient than scaling model parameters, with 7B models plus novel search outperforming 34B models.
A pre-activation regularizer seeds more affine regions near data in piecewise affine networks, increasing local region count and improving early training performance.
On five tabular security datasets at 10% labels, tuning only the classifier with Bayesian optimization recovers a median 86% of the gains from full joint SSL-classifier optimization.
Causality resolves trade-offs in trustworthy AI by treating them as invariance conflicts under different data-generating process changes.
citing papers explorer
-
Knowledge Beyond Language: Bridging the Gap in Multilingual Machine Unlearning Evaluation
New metrics KSS and KPS are introduced to evaluate multilingual machine unlearning quality and cross-language consistency in LLMs, addressing limitations of single-language evaluation protocols.
-
Participatory provenance as representational auditing for AI-mediated public consultation
A new auditing framework measures how much of the public's submitted viewpoints is lost in AI-generated consultation summaries, finding that official summaries represent the population worse than a random set of participants and that critical voices are most likely excluded.
-
Scaling Relations for Binary Black Hole Merger Times from Cosmological Initial Conditions
Cosmological MBHBs coalesce in ~1 Gyr with high eccentricities; scaling relations from 30 Griffin re-simulations link dynamical-friction, hardening and total times to galaxy and orbital properties.
-
Why SGD is not Brownian Motion: A New Perspective on Stochastic Dynamics
SGD is reformulated via a master equation from discrete updates, producing a discrete Fokker-Planck equation that predicts non-stationary variance growth proportional to learning rate in flat Hessian directions.
-
New Insight of Variance reduce in Zero-Order Hard-Thresholding: Mitigating Gradient Error and Expansivity Contradictions
A generalized variance-reduced ZO hard-thresholding algorithm removes prior limits on random directions for gradient estimates, yielding improved convergence rates under standard assumptions.
-
Query-efficient model evaluation using cached responses
DKPS-based methods predict new model benchmark scores using cached responses, matching baseline mean absolute error with substantially fewer queries and an offline query selection approach.
-
Linear Decision Tree Policies for Integer Linear Programs
Any fixed integer linear program with a finite feasible set can be answered by a precomputed linear decision tree using polynomially many arithmetic operations per cost query; a practical construction works on small instances.
-
Inference Scaling Laws: An Empirical Analysis of Compute-Optimal Inference for Problem-Solving with Language Models
Empirical analysis shows scaling inference compute via strategies like tree search can be more efficient than scaling model parameters, with 7B models plus novel search outperforming 34B models.
-
Region Seeding via Pre-Activation Regularization: A Geometric View of Piecewise Affine Neural Networks
A pre-activation regularizer seeds more affine regions near data in piecewise affine networks, increasing local region count and improving early training performance.
-
SemiScope: Disentangling Classifier Tuning and Joint Optimization in Semi-Supervised Security Classification
On five tabular security datasets at 10% labels, tuning only the classifier with Bayesian optimization recovers a median 86% of the gains from full joint SSL-classifier optimization.
-
Trustworthy AI Suffers from Invariance Conflicts and Causality is The Solution
Causality resolves trade-offs in trustworthy AI by treating them as invariance conflicts under different data-generating process changes.