paper summary
94 notes, newest first.
SAPO: Single-Rollout Autoregressive Policy Optimization for Agentic Reinforcement Learning
A new reinforcement learning method called Single-rollout Autoregressive Policy Optimization (SAPO) is proposed to address the memory overhead, computation efficiency, and performance issues of existing methods. SAPO…
Eureka: Task-Conditioned Meta-Agent Orchestration for Scientific Discovery
The paper introduces Eureka, a meta-agent system designed for scientific discovery tasks. Eureka forms specialized agents through planning, promotion, and minimal-sufficient compilation, adapting to recurring…
ORBITER: Conflict-Aware Decision-Making for Agentic Last-Mile Delivery
The paper introduces ORBITER, a system designed to make decisions for last-mile delivery by couriers. It models the courier service through decision points and uses language models (LLMs) to reason about individual…
rEDMRec: Distilling Large Language Model Reasoning into an Editable Experience Memory for Recommendation
The authors propose a system called rEDMRec that improves recommendation quality by compressing the reasoning of a large language model (LLM) into a compact, editable memory. This memory is maintained by an LLM memory…
Robust Risk Under Evolving Uncertainty: A Wasserstein Counterpart of the Entropic Value-at-Risk
The paper proposes a new risk measure called Wasserstein Entropic Value-at-Risk (WEVaR) for an agent learning its environment. The WEVaR is designed to account for the reachable catastrophes that the current entropic…
SPADE: Self-Play in Adaptive Synthetic Executable Environments
The paper introduces SPADE (Self-Play in Adaptive Synthetic Executable Environments), a framework that allows a language model to design and learn in its own self-generated, diverse, adaptive environments. This…
SPADE: Self-Play in Adaptive Synthetic Executable Environments
The paper introduces SPADE, a reinforcement learning framework that allows a language model to act as both an Environment Designer and a Reasoning Agent. The Environment Designer generates long-horizon training…
Chronocooked: A Benchmark for Implicit Interval Timing in Reinforcement Learning Agents
The paper introduces Chronocooked, a benchmark suite for studying how reinforcement learning (RL) agents make decisions over time. It's inspired by Overcooked and includes cooking scenarios that require temporal…
ReLTEx: Reliable LLM-based Taxonomy Expansion
The paper discusses a new framework called ReLTEx for improving the reliability of Large Language Model (LLM)-based taxonomy expansion. The framework combines LLM-driven candidate generation with structure-aware…
sLTN: Structural Logic Tensor Networks
The paper introduces sLTN, an extension of Logic Tensor Networks (LTN), which is a neurosymbolic framework that combines first-order logic with tensor operations for differentiable learning. The new version can handle…
BDH-CQ: In-Context Learning with Recurrent Latent Reasoning
The paper introduces BDH-CQ, a model that combines in-context learning and recurrent latent reasoning. The model updates its memory based on inputs presented during inference, solves queries through iterative…
Comparative Approaches to Agent Retrieval over Large Skill Libraries
The research compares two systems for deciding which skills to load from a large skill library when creating agents. One system combines lexical and dense-embedding retrieval, while the other uses a typed knowledge…
EnvACE: Internalizing Environment Dynamics via World Rehearsal for Agentic Reinforcement Learning
The researchers introduce EnvACE, a method for reinforcement learning that replaces external environment interactions during training with world rehearsal. This method allows the policy to internalize the relationship…
Agentic Reinforcement Learning with Observation-Calibrated Self-Distillation
Researchers propose Observation-Calibrated Self-Distillation (OCSD) to address a confounding issue in On-Policy Self-Distillation (OPSD), a method used to train large language model agents. OCSD aims to attribute…
Chained Recursive Language Models for Multi-Iteration Reasoning
The paper proposes a new architecture for large language models called Chained Recursive Language Models (Chained RLM). This architecture is designed to handle long context reasoning tasks that require extraction…
Chained Recursive Language Models for Multi-Iteration Reasoning
The paper proposes Chained Recursive Language Models (Chained RLM), an architecture for multi-iteration reasoning in large language models (LLMs). It addresses the difficulty in tasks requiring extraction, counting…
From Score Matrices to Football-Aware Match-State Simulation: An Auditable LLM Harness for Exact-Score Reranking
The paper presents an auditable information harness that combines dynamic Poisson-family models and large language models (LLMs) to forecast football scores. The system goes through four iterations, each improving on…
ORACLE: A Multi-Objective Reinforcement Learning-Based Analog Circuit Design Optimizer with Large Language Models-Guided Exploration
The paper presents ORACLE, a multi-objective reinforcement learning-based framework for optimizing analog circuit design. The framework addresses limitations of existing methods by using vector-valued learning and…
A foundation model of numerical intelligence with cross-disciplinary generalization
The paper introduces UNified In-Context Operator Networks (UNICON), a model that demonstrates numerical intelligence across various disciplines. It is capable of inferring predictive relations from graph-based examples…
MANTA: Multi-Agent Network Topology Adaptation for Self-Evolving Multi-Agent Systems
The paper introduces MANTA, a framework for adapting communication structures in multi-agent systems during inference time. This allows for more efficient problem solving by enabling agents to adjust their roles…
MemHarness: Memory Is Reconstructed, Not Replayed
Researchers propose MemHarness, a framework for enhancing large language models that allows them to adapt and reconstruct past experiences based on the current context, rather than simply replaying them verbatim as…
SkillRise: Agentic Reinforcement Learning for Cross-Task Skill Evolution
The paper introduces SkillRise, a reinforcement learning framework that allows large language model agents to learn and reuse skills across related yet distinct tasks. SkillRise improves Pass@1 performance on ALFWorld…
UrbanDS: A Graph-Guided LLM Multi-Agent System for Data-Intensive Urban Tasks
The paper presents UrbanDS, a multi-agent system that uses a graph to organize and analyze large-scale and heterogeneous urban data. The system includes a Data Profiling Agent, a Relation Agent, a Planner Agent…
ClinFusion: A Vision-Centric Multimodal LLM System for Holistic Medical Understanding
The paper introduces ClinFusion, a system designed for holistic medical understanding using multimodal large language models (MLLMs). ClinFusion addresses challenges in the medical domain by unifying diverse 2D and…
ClinFusion: A Vision-Centric Multimodal LLM System for Holistic Medical Understanding
The paper introduces ClinFusion, a multimodal large language model designed for holistic medical understanding. It can handle both 2D and 3D medical images and outperforms other open-source and proprietary models on…
Sparse Autoencoders Encode Both Concepts and Functions: The Downstream Geometry of Feature Effects
A study explores the inconsistencies in linking sparse autoencoders (SAEs) with model behavior, focusing on the geometry of changes in model logits caused by feature interventions. The authors propose an unsupervised…
Detecting LLM-Generated Tokens in Human--LLM Coauthored Text
The paper presents a new method for identifying parts of a text generated by Language Model (LLM) at the token level, improving on existing document-level methods. The method is based on smoothing adjacent token scores…
Expert Behavior Prior Reinforcement Learning
The paper proposes a new method called Expert Behavior Prior (EBP) for behavior prior reinforcement learning (BPRL). EBP aims to improve the effectiveness of policy priors by generating expert policy priors directly…
MIRROR: Learning from the Other View for Multi-Modal Reasoning
The paper presents MIRROR, a method to improve multi-modal reasoning in AI models by learning from different views (text, image, combined text and image) of the same problems. The study reveals that vision-language…
Autonomous Collaborative Learning Among an Ensemble of Tsetlin Machines with Consensus-Based Inference
The researchers propose a decentralized collaborative learning paradigm for an ensemble of Tsetlin Machines (TMs) using consensus-based inference. Each agent has its own private TM model, and they combine individual…
Decentralized Online Riemannian Optimization for Strongly Geodesically Convex Functions
The authors study decentralized online optimization for specific types of functions (strongly geodesically convex) on Riemannian manifolds with certain curvature properties. They propose a new network-error analysis to…
Instance Hardness-Based Relevance for Imbalanced Regression
A new approach called Instance Hardness-based relevance function (InHaR) is proposed to identify rare instances in regression problems, particularly useful for complex scenarios like bimodal distributions where…
Instance Hardness-Based Relevance for Imbalanced Regression
The paper proposes a new relevance function called Instance Hardness-based relevance (InHaR) for identifying rare instances in imbalanced regression problems. This function takes into account the learning difficulty of…
Local Stability and Gaussian Smoothing of Quantized Neural Networks
The authors investigate using Gaussian averaging as a smooth alternative for quantized neural networks. They provide bounds on the difference between the original function (f) and the smoothed version (g), linking it…
Local Stability and Gaussian Smoothing of Quantized Neural Networks
The authors study the use of Gaussian averaging as a smoother alternative for quantized neural networks. They derive a bound on the difference between the original function and the smoothed version under certain…
On Optimization Complexity of Second-Order Certified Unlearning
The authors investigate the complexity of removing trained data from machine learning models, focusing on certified unlearning and optimization accuracy. They propose new bounds on the distance between initial and…
Statistical Inference for Rank Allocation in Low-Rank Adaptation
In this paper, the authors propose StatLoRA, a statistical inference-based rank allocation method for low-rank adaptation (LoRA) in large language models. The proposed method uses estimated p-values to determine which…
Statistical Inference for Rank Allocation in Low-Rank Adaptation
Researchers propose StatLoRA, a statistical inference-based method for rank allocation in low-rank adaptation (LoRA). This method aims to balance efficiency, expressiveness, and generalization by using estimated…
Variance-reduced Domain Adaptation using Paired Sampling
The paper proposes Paired Sampling for Domain Adaptation (PSDA), a new stochastic variance reduction technique designed for distribution-matching frameworks in unsupervised domain adaptation (UDA) that suffer from high…
H$^2$SD: Hybrid Hindsight Self-Distillation
The paper introduces H$^2$SD, a framework for reinforcement learning with verifiable rewards (RLVR) that addresses the limitations of previous methods. H$^2$SD uses a hybrid approach to self-distillation, where it…
Rationale-Guided Knowledge Distillation for Cross-Lingual Stance Detection
The authors propose a method called rationale-guided knowledge distillation for cross-lingual stance detection. This technique aims to identify attitudes in texts towards a target, and helps when there is a lack of…
AdaHome: An Adaptive Smart Home Assistant using Local Small Language Models
A paper titled 'AdaHome: An Adaptive Smart Home Assistant using Local Small Language Models' presents a new smart home assistant system designed to work with locally deployed small language models in…
Contextualized Early Detection of Online Firestorms: A Sequential LLM-Based Approach
The paper introduces an LLM-based system to detect online firestorms (rapid escalations of negative user-generated content) on Reddit. The system operates in two modes: retrospective thread classification and early…
Latent Trajectory Discrimination for AI-Generated Text Detection
In this paper, the authors propose a new approach for detecting AI-generated text called Geometric Trajectory and Contrastive Learning (GTCL). Unlike existing methods that focus on aggregate statistics or global…
Reachability-Aware Pretraining for Efficient Target-Oriented Path Exploration in Temporal Knowledge Graph Reasoning
The authors propose a new method called RAPTOR for improving the efficiency of target-oriented path exploration in temporal knowledge graphs (TKG). The method injects a reachability-aware inductive bias to the agent…
Show Me How You Reason and I'll Tell You Who You Are: Reasoning Graphs for Robust LLM Authorship Attribution
The paper proposes a new method for detecting text generated by large language models (LLMs) and attributing authorship to them. The method goes beyond just analyzing surface-level linguistic features and instead…
Memory as a Controlled Process: Learned Adaptive Memory Management for LLM Agents
The paper proposes a new framework called Memory as a Controlled Process (MemCon) for large language model (LLM) agents that use external memory systems. MemCon learns an online policy to adaptively decide when, what…
Unleashing Multimodal Large Language Models for Training-free HOI Detection in the Wild
The paper presents a new framework called AgentHOI that uses multimodal large language models (MLLMs) to detect human-object interactions (HOIs) in complex scenes without the need for any HOI data during training. The…
Do AI Agents Know When a Task Is Simple? Toward Complexity-Aware Reasoning and Execution
Researchers propose E3, a method to help large language model agents estimate the complexity of tasks and reduce unnecessary work. The proposed method reduces cost, tokens, and inspected files by significant margins on…
Tracing Agentic Failure from the Flow of Success
The paper proposes a new method called OAT (One-Class Attribution for Tracing) for identifying errors in language models without relying on expensive or difficult-to-collect data. This method is faster and more…
LoRA-Based Cascaded Multimodal Fusion for Action Recognition in Medical Training Environments
The paper introduces a new framework for recognizing actions and activities in medical training environments using a cascaded Low-Rank Adaptation (LoRA)-based multimodal fusion approach. This method allows for the…
Cognitive-structured Multimodal Agent for Multimodal Understanding, Generation, and Editing
Researchers propose a new type of multimodal agent called the Cognitive-structured Multimodal Agent (CMA) that can understand, generate, and edit visual and textual information. The agent externalizes visual…
SLORR: Simple and Efficient In-Training Low-Rank Regularization
The paper introduces SLORR, a simple and efficient low-rank regularization method for compressing neural networks during training. It addresses the limitations of existing methods by being stateless…
SLORR: Simple and Efficient In-Training Low-Rank Regularization
The paper introduces SLORR, a simple and efficient in-training low-rank regularization framework for neural networks. It aims to improve compressibility without significant accuracy loss, while avoiding the limitations…
SLORR: Simple and Efficient In-Training Low-Rank Regularization
The paper introduces SLORR, a simple and efficient in-training low-rank regularization framework for compressing neural networks without significant accuracy loss or relying on stateful cached quantities. The authors…
WebSwarm: Recursive Multi-Agent Orchestration for Deep-and-Wide Web Search
The paper proposes a new framework called WebSwarm for multi-agent search systems. This system aims to improve deep-and-wide web searches by dynamically creating search nodes that can delegate tasks recursively and…
WebSwarm: Recursive Multi-Agent Orchestration for Deep-and-Wide Web Search
The paper presents a new framework called WebSwarm for web search using multiple agents. This system aims to overcome the limitations of single agents by simultaneously handling depth and coverage, improving recursive…
ORCAID: Oblique Rule-Based Continuous-Action Interpretation for Deep RL Policies
The paper introduces ORCAID, a method for extracting interpretable rule-based policies from reinforcement learning (RL) agents in complex environments with continuous action spaces. The method involves training an…
Single-Rollout Asynchronous Optimization for Agentic Reinforcement Learning
The paper proposes a new method called Single-rollout Asynchronous Optimization (SAO) for Reinforcement Learning (RL) in post-training large language models (LLMs). SAO aims to improve stability and task effectiveness…
Single-Rollout Asynchronous Optimization for Agentic Reinforcement Learning
The paper introduces Single-rollout Asynchronous Optimization (SAO), a new approach to reinforcement learning (RL) aimed at post-training large language models (LLMs). SAO addresses stability and off-policy challenges…
Single-Rollout Asynchronous Optimization for Agentic Reinforcement Learning
In this paper, the authors introduce Single-rollout Asynchronous Optimization (SAO), a new approach to reinforcement learning (RL) for post-training large language models (LLMs). SAO aims to improve training stability…
Danus: Orchestrating Mathematical Reasoning Agents with Fact-Graph Memory
The paper introduces Danus, a system designed to effectively coordinate and scale mathematical reasoning agents by using a shared fact graph as a global memory-management mechanism. The system consists of a main agent…
Information Gain-based Rollout Policy Optimization: An Adaptive Tree-Structured Rollout Approach for Multi-Turn LLM Agents
The paper presents Information Gain-based Rollout Policy Optimization (IGRPO), a new framework for improving large language model agents in long-horizon search tasks. IGRPO aims to allocate rollout budget more…
TacReasoner: A Dynamic Tactile-Language Framework for Interactive Reasoning in Real-World Scenarios
The paper proposes TacReasoner, a new framework that enhances the perception and representation of dynamic tactile signals for intelligent systems. It addresses two challenges: insufficient modeling of such signals and…
Unified Audio Intelligence Without Regressing on Text Intelligence
Researchers have developed a unified language model called Audex, which combines audio and text intelligence. The model is built on Nemotron-Cascade-2-30B-A3B, a strong text-only MoE LLM (Language Model). Audex uses a…
Weak-to-Strong Generalization via Direct On-Policy Distillation
Researchers propose a method called Direct On-Policy Distillation (Direct-OPD) for improving stronger language models by leveraging the knowledge gained from reinforcement learning on smaller, weaker models. This…
Learning to Move Before Learning to Do: Task-Agnostic pretraining for VLAs
Researchers propose Task-Agnostic Pretraining (TAP), a two-stage framework to train Vision-Language-Action (VLA) models. TAP first learns transferable motor priors from unlabeled interaction data using self-supervised…
Neuron-Aware Data Selection for Annotation-Free LLM Self-Distillation
The paper proposes a new method called Neuron On-Policy Self-Distillation (Neuron-OPSD) for annotation-free self-distillation of large language models. This method uses the internal neuron activations to guide data…
Neuron-Aware Data Selection for Annotation-Free LLM Self-Distillation
In this paper, the authors propose a new method called Neuron On-Policy Self-Distillation (Neuron-OPSD) for annotation-free self-distillation of large language models. This method uses internal neuron activations to…
QFedAgent: Quantum-Enhanced Personalized Federated Learning for Multi-Agent Activity Recognition
The paper proposes a new method called QFedAgent for multi-agent activity recognition in privacy-sensitive robotic sensing applications. The method uses a hybrid quantum-classical personalized Federated Learning…
CausalMix: Data Mixture as Causal Inference for Language Model Training
The paper presents CausalMix, a method that addresses the limitation of existing data mixing methods in Large Language Model (LLM) training by casting it as a causal inference problem. By using causal modeling to…
Human-Machine Collaboration on Generative Meta-Learning: Model and Algorithm
The research paper introduces a new framework called Generative Meta-Learning with Human Feedback (GMHF). This framework aims to improve machine learning models by bridging the gap between training and target domains…
Right in the Right Way: LM Training with Verifiable Rewards and Human Demonstrations
Researchers propose an adversarial generator-discriminator framework to train language models (LM) for tasks with well-defined success metrics, like code generation and mathematical reasoning. The new approach improves…
ECHO: Prune to act, trace to learn with selective turn memory in agentic RL
The paper proposes ECHO, a selective turn-memory framework for long-horizon language agents. ECHO aims to address the issues of history collapse and traceable learning by compressing completed environment turns into…
MECoBench: A Systematic Study of Multimodal Agent Collaboration in Embodied Environments
The paper introduces MECoBench, a new system for testing the collaboration abilities of multimodal large language models in visually grounded environments. The authors conduct extensive experiments with various MLLMs…
TRIAGE: Role-Typed Credit Assignment for Agentic Reinforcement Learning
The paper proposes a new framework called TRIAGE for credit assignment in agentic reinforcement learning. This framework adds a semantic role axis to outcome credit, aiming to correct the issues of standard GRPO where…
AgentsNet: Coordination and Collaborative Reasoning in Multi-Agent LLMs
The paper proposes AgentsNet, a new benchmark for multi-agent reasoning. This benchmark is designed to measure the ability of multi-agent systems to collaborate effectively and leverage their network topology. The…
C$^{2}$R: Cross-sample Consistency Regularization Mitigates Feature Splitting and Absorption in Sparse Autoencoders
The paper introduces C$^2$R, a method to improve Sparse Autoencoders (SAEs) used for interpreting large language models. The main challenge with SAEs is scaling to large dictionaries, which leads to issues like feature…
DOPD: Dual On-policy Distillation
The paper proposes a new method called Dual On-policy Distillation (DOPD) for transferring knowledge from a teacher model to a student model in on-policy distillation. DOPD aims to address the issue of 'privilege…
Estimating Commonsense Plausibility through Semantic Shifts
The paper proposes a new method called ComPaSS for evaluating the plausibility of language by measuring changes in semantics when adding common sense-related information to sentences. The method performs better than…
RETRIEVALFORMER: TRANSFORMER-QUALITY RECOMMENDATIONS WITH EFFICIENT ANN RETRIEVAL AND COLD-START RESILIENCE
The researchers propose RetrievalFormer, a transformer-based recommendation system that combines high accuracy with efficient Approximate Nearest Neighbor (ANN) retrieval. This system can recommend unseen items during…
Joint Learning of Experiential Rules and Policies for Large Language Model Agents
A research paper proposes a method called Joint Learning of Experiential Rules and Policies (JERP) to improve the effectiveness of Large Language Model Agents in multi-step interactive environments. The method updates…
ShareLock: A Stealthy Multi-Tool Threshold Poisoning Attack Against MCP
Researchers introduce a new method called ShareLock, a multi-tool threshold poisoning framework that uses Shamir's threshold scheme to inject malicious instructions into LLM-servers through MCP. The method disperses…
ShareLock: A Stealthy Multi-Tool Threshold Poisoning Attack Against MCP
The paper introduces a new method called ShareLock, which is a multi-tool threshold poisoning framework that can exploit the interactions between Language Model servers and external tools. This method is designed to…
Explainable Control Framework (XCF) based on Fuzzy Model-Agnostic Explanation and LLM Agent-Supported Interface
The paper introduces an Explainable Control Framework (XCF) for understanding complex controllers' behavior in various scenarios. It offers three main contributions: the XCF, a model-agnostic explanation method called…
FORCE: Efficient VLA Reinforcement Fine-Tuning via Value-Calibrated Warm-up and Self-Distillation
The paper introduces FORCE, a framework designed to improve the efficiency of reinforcement learning fine-tuning for Vision-Language-Action (VLA) models. The framework consists of three stages: Value-Calibrated…
Multi-Agent Goal Recognition with Team- and Goal-Conditioned Reinforcement Learning and Factorized Branch-and-Bound
The paper presents Multi-Agent Goal Recognition with Team- and Goal-Conditioned Reinforcement Learning and Factorized Branch-and-Bound (MAGR-BB), a method to infer which agents in a multi-agent system are working…
On-Policy Self-Distillation with Sampled Demonstrations Reduces Output Diversity
The paper discusses a method called On-policy Self-Distillation, which uses a single model as both teacher and student to achieve high accuracy in pass@1. However, it is found that this method can lead to a decrease in…
Weave of Formal Thought
The paper introduces Weave of Formal Thought (WoFT), a new paradigm for large language models that combines syntactic validation with learned structural representations. The proposed approach includes a formal engine…
ASALT: Adaptive State Alignment for Lateral Transfer in Multi-agent Reinforcement Learning
The paper introduces a new method called ASALT for multi-agent reinforcement learning (MARL). This method is designed to handle situations where the state-space dimensionalities between source and target domains don't…
ASALT: Adaptive State Alignment for Lateral Transfer in Multi-agent Reinforcement Learning
The paper introduces ASALT, a method for multi-agent reinforcement learning (MARL) that allows for knowledge transfer between source and target domains even if the state space dimensions do not match. ASALT uses…
FlowPipe: LLM-Enhanced Conditional Generative Flow Networks for Data Preparation Pipeline Construction
Researchers propose a new method called FlowPipe to automatically construct data preparation pipelines for machine learning. The proposed method uses Conditional Generative Flow Networks (C-GFlowNets) and Deep Semantic…
LaGO: Latent Action Guidance for Online Reinforcement Learning
The paper 'LaGO: Latent Action Guidance for Online Reinforcement Learning' proposes a new framework called LaGO. This framework uses a pre-trained language model (LLM) as a guide to help with online policy optimization…
ScaleToT: Generalizing Structured LLM Reasoning for Billion-Scale Low-Activity User Modeling
The paper presents ScaleToT, a method that improves user modeling for low-activity users by learning structured reasoning from a small subset of Large Language Models (LLMs). It constructs typed user-state chains using…
