All papers

Proposes an optimization method to discover undisclosed or harmful generation concepts embedded within diffusion model adaptation weights.

The paper introduces a diagnostic pipeline to identify low-confidence patient subgroups and detect harmful covariate shifts in clinical prediction models.

The authors present IDGuardian, an adversarial perturbation framework designed to defend facial images against unauthorized personalized diffusion models.

The paper introduces an authorization system that reuses validated consequence evidence across agent execution branches while maintaining separate live authority.

Proposes a decentralized coordination mechanism using commitment tracking to preserve clear accountability across multi-agent delegation chains.

Proposes a lossy compression framework for federated learning that balances user privacy against model utility on resource-constrained devices.

The paper introduces ZoomCap, an adaptive visual zooming and evidence-routing framework to improve fine-grained image captioning supervision.

Evaluates the security of streaming image classifiers to edge devices against physical cold-boot memory extraction attacks.

This paper presents a benchmark and link inference attack framework to evaluate edge unlearning methods in graph neural networks.

This paper presents a pathwise conformal prediction method that guarantees valid coverage across adaptive early-exit stopping policies without multi-exit score inflation.

Introduces a metric to measure how non-linearly features learned by sparse autoencoders interact during downstream model behaviors.

Presents an exact machine unlearning method for federated socially fair k-means clustering using separate group-level local summaries.

This paper introduces a training-free safety filter that evaluates scene context to constrain generated humanoid motion trajectories.

This paper presents a feasibility-aware safety filter that amortizes minimal-intervention constraints using a feedforward neural network in reinforcement learning.

This paper develops a unified Bellman operator to optimize task performance and enforce constraints within safety-critical reinforcement learning.

The paper proposes a credal machine learning framework that models epistemic uncertainty with probability distribution sets for risk-averse decision-making.

The paper establishes and validates a set of meta-requirements to guide the design of frameworks for auditing AI system robustness.

The paper argues that full robotic value alignment is impossible and examines how to manage acceptable misalignment in autonomous systems.

The paper presents GENDIS, a multilingual dataset of YouTube comments from India, Kenya, and Nigeria annotated for gender-related discourse and stance.

This paper evaluates whether attention-graph curvature features can help verifiers detect errors in mathematical reasoning steps.

The paper presents a byte-local text watermarking method that scores outputs without invoking tokenizers during detection.

Studies uncertainty estimation for abstract meaning representation parsers using top-k beam candidates and prediction rejection ratios.

Introduces an immutable, provenance-aware state representation framework with deterministic replay and audit logging for robotic systems.

Proposes a conformal prediction framework that learns stochastic groupings to improve subgroup coverage reliability without inflating overall prediction sets.

Tests whether priming annotators with scenario vignettes and micro-narratives improves the operationalization of lived experience in subjective annotation.

The paper demonstrates an on-chip power side-channel attack reconstructing color ImageNet inputs from machine learning accelerators on shared multi-tenant FPGAs.

The paper introduces a black-box method to extract training sequences from genomic language models using output sampling and sequence alignment.

This paper presents an interface framework applying visual transformations and adversarial attacks to create assignment text resistant to machine extraction.

This paper proposes an aggregator-agnostic watermarking scheme to embed persistent watermarks into models during federated learning despite robust aggregators.

This paper introduces a server-side defense against active label inference attacks in vertical federated learning using embedding asymmetry signals.

This paper proposes an external memory-based framework to unlearn specific identities from trained models when training data is absent.

The paper proposes certified adversarial detection using Lipschitz-margin analysis to detect perturbations rather than forcing invariant predictions.

This paper proposes a counterfactual certification framework to bound performance loss during sequential machine unlearning requests.

The authors analyze robustly reliable learning for regression under data corruptions and propose an algorithm to compute optimal prediction intervals for Lipschitz predictors.

This paper presents a psychometric framework to test whether automated sentiment classification remains equivalent across different social media platforms.

This paper measures whether ecological alignment in large language models survives under pressure using a calibrated pretraining ladder.

This paper introduces a data-flow graph representation and mutation testing technique to uncover bugs in deep learning frameworks.

This paper proposes a hierarchy of sparse second-order cone programming relaxations for certified neural network robustness.

The paper presents a framework using random noise and Sigmoid function analysis to detect directional adversarial attacks on medical imaging models.

The paper develops methods to evaluate and recalibrate binary predictors using noisy, multi-annotator labels without requiring clean ground truth.

The paper proposes a sequential conformal prediction algorithm that groups calibration scores by state similarity rather than pure temporal recency.

Proposes a multimodal reasoning framework to avoid biased role attributions in underspecified or ambiguous social memes.

The paper analyzes physical recursive self-improvement in robotics, proposing a multi-level evaluation framework and reviewing existing adaptation methods.

The authors analyze readout scaling in sparse autoencoders and propose canonicalized coordinates to improve the recovery of underlying generative factors.

This paper investigates theoretical limits and conditions for verifying reachability policies in unknown Markov decision processes using sample access.

The paper presents a coding method that embeds hidden communication into an agent's policy randomness without altering its observable action distribution.

This paper models preference model alignment as a Stackelberg game to identify comparisons where evaluators are prone to error.

The paper analyzes theoretical computational limits of sequential sampling circuits in autoregressive language model token generation.

The paper specifies a verifier-governed protocol for panvascular research and benchmarks a baseline model on intervention imaging tasks.

This paper establishes minimax rates for estimating language model semantic entropy using verdicts from weak pairwise equivalence judges.

The paper introduces a neurosymbolic world model that learns interpretable concepts from temporal logic constraints for agent planning.

Proposes input loss curvature as a sampling-free metric to detect the onset of training data memorization in diffusion models.

Introduces a welfare-aware objective to allocate limited human feedback across diverse individuals when training pluralistic alignment models.

This paper introduces a multi-hop reasoning benchmark to evaluate fine-grained, explainable hate speech analysis in Persian.

The paper proposes a hierarchical two-loop optimization framework that uses residual errors to trigger active evidence acquisition during inference.

The paper presents an online Thompson Sampling bandit policy to select classifiers that balance predictive accuracy and subgroup fairness under delayed feedback.

This paper establishes theoretical limits on what forms of annotator disagreement can be identified from pairwise preference feedback data.

Proposes conformal calibration to weight criterion-level uncertainty when using rubric-based language model judges for reinforcement learning.

The paper introduces a parameter decomposition method to interpret the mechanistic subcomponents of language model weights.

The paper examines AI-generated text detection in scientific papers by combining visual formatting, PDF text, and LaTeX representations.