AI Research

146 items

RESEARCHarXiv CS.LG·29d ago

Path-Based Gradient Boosting for Graph-Level Prediction

We propose PathBoost, a gradient tree boosting method for graph-level classification and regression, which learns discriminative path-based features directly from the input graph structure. This method introduces adaptations for binary classification, incorporates multiple node and edge attributes, and automatically selects anchor nodes, outperforming or matching graph neural networks and graph kernel approaches on several benchmark datasets.

gradient boosting Graph Neural Networks machine learning graph theory

ARTICLEMIT Tech Review AI·4/13/2026

You have no choice in reading this article—maybe

Uri Maoz, during his PhD, delved into computational neuroscience, studying how the brain instructs arm movements and subsequently perceives that motion. His research explored the intricate mechanisms of human motor control and sensory perception.

computational neuroscience perception brain motor control

RESEARCHarXiv CS.CL·8d ago

lmfaoooo at SemEval-2026 Task 1: Humor Is an Audience. Preference Modeling for Constrained Humor Generation

This paper describes a system for SemEval-2026 Task-1, which focuses on constrained humor generation. The approach uses a

evaluation Natural Language Processing humor generation AI Research

RESEARCHarXiv CS.AI·29d ago

Belief or Circuitry? Causal Evidence for In-Context Graph Learning

This paper investigates how LLMs learn in-context, using a graph random-walk task to explore whether they pattern-match or infer latent structure. It reveals that neither account alone is sufficient, presenting evidence of simultaneous encoding of graph topologies and causal interventions.

LLMs learning interpretability graph learning

RESEARCHarXiv CS.CL·26d ago

Merging Methods for Multilingual Knowledge Editing for Large Language Models: An Empirical Odyssey

This paper investigates the effectiveness of vector merging methods for multilingual knowledge editing (MKE) in Large Language Models, focusing on reducing interference between language-specific edits. Evaluating six merging variants across two LLMs, two editing methods, and 12 languages on the MzsRE benchmark, it finds vector summation with shared covariance to be the most reliable overall strategy.

multilingual LLMs Natural Language Processing Vector Merging Knowledge Editing

RESEARCHarXiv CS.CL·28d ago

HEBATRON: A Hebrew-Specialized Open-Weight Mixture-of-Experts Language Model

Hebatron is a Hebrew-specialized open-weight large language model built on NVIDIA's Nemotron-3 Mixture-of-Experts (MoE) architecture. It achieves a 73.8% Hebrew reasoning average, outperforming competitors and offering significantly higher inference throughput by activating fewer parameters per pass.

language models NVIDIA AI Hebrew AI Mixture of Experts

RESEARCHarXiv CS.AI·28d ago

EVOCHAMBER: Test-Time Co-evolution of Multi-Agent System at Individual, Team, and Population Scales

EVOCHAMBER presents a training-free framework that instantiates test-time evolution at three levels over a coevolving agent pool, distinguishing it from single-agent approaches. It features CODREAM, a post-task protocol for collaborative reflection and asymmetric knowledge routing after team failures or disagreements.

Evolutionary AI machine learning multi-agent systems Collaboration

RESEARCHarXiv CS.CL·27d ago

Bridging the Missing-Modality Gap: Improving Text-Only Calibration of Vision Language Models

Vision-language models (VLMs) experience significant accuracy drops and severe miscalibration when operating with text-only inputs, even with preserved semantic information. The Latent Imagination Module (LIM) is proposed to predict imagined latent embeddings from text, improving accuracy and reducing calibration error in missing-image scenarios.

Miscalibration Vision-Language Models Latent Imagination Text-Only Inputs

RESEARCHarXiv CS.CL·28d ago

Sampling More, Getting Less: Calibration is the Diversity Bottleneck in LLMs

This research addresses the lack of diversity in LLM outputs, attributing it to how models allocate probability mass across valid and invalid continuations during decoding. It introduces a validity-diversity framework that decomposes the problem into two complementary forms of miscalibration: order calibration and shape calibration.

Calibration diversity LLMs decoding

RESEARCHarXiv CS.CL·27d ago

BoostTaxo: Zero-Shot Taxonomy Induction via Boosting-Style Agentic Reasoning and Constraint-Aware Calibration

BoostTaxo introduces a novel boosting-style LLM framework designed for zero-shot taxonomy induction, aiming to overcome limitations in generalization and efficiency of existing methods. It refines taxonomy construction through a coarse-to-fine parent identification process, leveraging retrieval-augmented definition refinement and hybrid candidate selection.

Taxonomy induction Semantic hierarchies AI Research LLM

RESEARCHarXiv CS.LG·12d ago

Balancing Multimodal Learning through Label Space Reshaping

The paper addresses modality imbalance in multimodal learning, where some modalities dominate optimization while others remain undertrained. It proposes that this discrepancy stems from differing mapping difficulties between modality-specific feature space and the shared label space, introducing BMLR to equalize this difficulty.

multimodal learning Optimization learning machine learning

RESEARCHarXiv CS.LG·12d ago

Continuity and Ordinality Matter: Constraining Time Series Tokens for Effective Time Series Analysis with Large Language Models

This paper introduces COM (Continuity and Ordinality Matter), a strategy that integrates geometric constraints into both the initialization and training stages of token-based time series large language models (TS-LLMs). The research demonstrates that preserving continuity and ordinality in time series token embeddings significantly improves model performance and generalizability.

machine learning Tokenization large language models Time Series Analysis

RESEARCHarXiv CS.CL·19d ago

Does Slightly Mean Somewhat? Measuring Vague Intensity Words in LLM Numeric Actions

This study investigates how large language models (LLMs), specifically Claude Haiku, interpret vague intensity words when producing numeric actions. The research reveals that the model compresses 10 intensity words into 5 distinct median outputs and is influenced by the current system state.

LLMs language interpretation numeric actions NLP

RESEARCHarXiv CS.CL·7d ago

On the Persistent Effects of Lexicality in Large Language Mod

This work investigates the persistent effect of lexical overlap, rather than semantic content, on representations extracted from large language models (LLMs) and its implications. The authors find that lexical influence extends across model depths, architectures, and training regimes, even in models trained for semantic similarity.

LLMs lexicality NLP semantic analysis

RESEARCHarXiv CS.CL·15d ago

EchoDistill:Alignment Noisy-to-Clean Self-Distillation for Robust Audio LLMs

EchoDistill is an alignment-based self-distillation framework designed to make Audio Large Language Models (ALLMs) robust to real-world noise. It leverages a frozen clean-audio teacher to guide an inference-time noisy-audio student, optimizing responses via group-relative policy optimization and token-level consistency.

robustness Audio LLMs machine learning Self-Distillation

RESEARCHarXiv CS.AI·16d ago

PathCal: State-Aware Reflection-Marker Calibration for Efficient Reasoning

This research paper introduces 'PathCal', investigating the distinct functional roles and timing of reflection markers in Large Reasoning Language Models' Chain-of-Thought trajectories. It reveals that markers like 'wait' or 'but' differ significantly in their impact on accuracy and generation length, challenging previous coarse-grained approaches.

Natural Language Processing Chain-of-Thought Reasoning large language models

RESEARCHarXiv CS.CL·16d ago

Graph Alignment Topology as an Inductive Bias for Grounding Detection

Large Language Models (LLMs) are optimized for plausible continuations rather than explicitly verifying if generated propositions are entailed by source documents, limiting their use in critical domains. This research proposes leveraging alignment topology as an inductive bias by constructing aligned bipartite graphs between reference information and LLM outputs, then training a Graph Neural Network (GNN).

LLMs hallucination grounding detection GNNs

RESEARCHarXiv CS.CL·7d ago

Linear Probes Detect Task Format, Not Reasoning Mode in Language Model Hidden States

This paper reveals that linear probes, often used to identify distinct reasoning representations in LLM hidden states, actually detect task format rather than reasoning modes. High accuracy observed on benchmarks with Qwen3-14B vanished when controlling for format variables, suggesting largely shared reasoning not functionally linked to hidden state geometry.

Benchmarking Natural Language Processing Model Analysis AI Research

RESEARCHarXiv CS.AI·15d ago

In Search of the Ingredients of Open-Endedness: Replicating Picbreeder with Large Vision-Language Models

This research explores AI's capacity for open-ended discovery in creative production by replicating Picbreeder with Vision-Language Models. It observes clear qualitative differences between AI-generated outputs and historical human baselines, attempting to characterize them.

Open-Ended Learning Vision-Language Models Evolutionary AI AI Research

RESEARCHarXiv CS.AI·16d ago

NeuroNL2LTL: A Neurosymbolic Framework for Natural Language Translation of Linear Temporal Logic

NeuroNL2LTL is a neurosymbolic architecture that unifies learned translation with formal verification to translate natural language into Linear Temporal Logic. It employs verifier-in-the-loop training, where verification outcomes serve as reward signals for reinforcement learning, optimizing for formal correctness.

reinforcement learning Neurosymbolic AI Formal verification Natural Language Processing