HDATF
English
Start a conversation
Menu

Paper Archive

Every archived paper with its summary, highest number first.

Papers 1,430Archive updated Research MonitoringBenchmarksRSS feed
  1. 231Computer VisionarXiv:2605.26115

    TriSplat: Simulation-Ready Feed-Forward 3D Scene Reconstruction

    • It proposes a structure that uses oriented triangles instead of Gaussians as the primitive, so the rendered output is already a mesh.
    • It derives normals from the predicted point map and refines them with an image-conditioned U-Net to keep triangle orientation aligned with geometry.
    • Mono-normal bootstrapping plus validity-aware masking stabilizes early training, and opacity and blur scheduling gradually sharpen the surface.
    • It achieves better geometry fidelity and competitive novel-view synthesis quality than Gaussian feed-forward baselines on RealEstate10K and DL3DV.
    • Zero-shot ScanNet evaluation confirms cross-dataset generalization.
    • It can be used directly in physics engines, collision detectors, and standard rendering pipelines without post-processing such as TSDF fusion.
    AbstractPDF
  2. 230Machine LearningarXiv:2605.23872

    Training-Free Looped Transformers

    • This method delivered consistent gains on knowledge-intensive benchmarks; for example, Qwen3-4B-Instruct improved by 2.64 percentage points on MMLU-Pro.
    • Unlike simple looping, which collapses as more iterations are added, the damped RK method remains stable or keeps improving as K increases, as shown in Figure 3.
    • The extra cost is proportional to the number of layers passed through, but because only a small window, such as four layers, is looped, the total runtime overhead is often only about 20 percent for a 32-layer model when the full decode loop is used; the authors also introduce a bypass mode that loops only during initial prompt processing, which makes token-generation overhead close to zero.
    AbstractPDF
  3. 229Computer VisionarXiv:2605.23903

    Geo-Align: Video Generation Alignment via Metric Geometry Reward

    • Standard training losses compare pixels or abstract features, and they do not understand physical units such as meters or degrees.
    • By using a 3D evaluator, the model learns to interpret camera motion in physical units, which resolves the scale drift that has long been a problem in video synthesis.
    • This RL framework removes the need for hard-to-obtain synchronized multiview video, allowing the model to learn from a wide range of real-world distributions.
    AbstractPDF
  4. 228Machine LearningarXiv:2605.23901

    LLMs as Noisy Channels: A Shannon Perspective on Model Capacity and Scaling Laws

    • Irreducible noise, e, corresponds to the constant term in Chinchilla-style laws and represents a fundamental structural limit or system entropy that cannot be eliminated by scaling.
    • Strategic scaling: the presence of a U-shaped curve suggests that there is an optimal sweet spot for scaling N and D at a given noise level. Blind scaling is not always the answer, and developers should consider the signal-to-noise ratio of their training setup.
    • Noise awareness: exponent analysis shows that data noise often grows faster than the learning signal, with delta greater than beta. This underscores the critical importance of data curation and noise mitigation. If your training data is noisy, simply adding more of it can eventually backfire.
    AbstractPDF
  5. 227ResearcharXiv:2605.23463

    StepAudio 2.5 Technical Report

    • ASR accuracy: The model achieved an average character error rate (CER) of 2.97% on Chinese benchmarks and a word error rate (WER) of 3.68% on English benchmarks. Its 32K context window makes it especially strong on long-form audio.
    • Decoding efficiency: Thanks to the MTP-5 mechanism, the model achieved a real-time factor (RTF) of 0.0053, making it faster than several specialized ASR systems. The acceptance rate of MTP tokens shows a consistent decay factor of about 0.9, validating the choice of five preceding branches.
    • TTS preference: In arena-style evaluation, StepAudio 2.5 TTS achieved an overall win rate of 67.6% against existing competitors such as ElevenLabs and Gemini Flash TTS.
    AbstractPDF
  6. 226Computer VisionarXiv:2605.23888

    GenRecon: Bridging Generative Priors for Multi-View 3D Scene Reconstruction

    • These chunk-wise predictions are merged back into the global grid by averaging the values in overlapping regions.
    • Scalability: chunk-based generation allows generative models, which were previously limited to single objects, to be applied to large multi-room scenes.
    • Geometric control: spatially grounded conditioning keeps the model's hallucinations anchored to the physical reality defined by camera pose and input images.
    AbstractPDF
  7. 225AI / GeneralarXiv:2605.23904

    SkillOpt: Executive Strategy for Self-Evolving Agent Skills

    • A skill optimized for a large model, such as GPT-4o, can be successfully applied to a smaller model, such as GPT-4o-mini, and the smaller model often captures much of the improvement, which suggests that the skill provides general procedural knowledge missing from the smaller model's weights.
    • A skill learned in one execution environment, such as the default chat interface, often remains valid when transferred to a more complex agent harness, such as a code execution sandbox, which indicates that the learned rule captures task-level logic rather than exploiting quirks of the training environment.
    • Procedures learned on one dataset, such as OlympiadBench, also improved performance on a related but distinct dataset, such as Omni-MATH, further showing that SkillOpt does not merely overfit to a specific set of training examples.
    AbstractPDF
  9. 223Computer VisionarXiv:2605.23902

    PiD: Fast and High-Resolution Latent Decoding with Pixel Diffusion

    • The pixel diffusion prior is a 1.3B-parameter Transformer optimized for high-resolution synthesis with a rectified-flow objective.
    • The latent adapter is a convolutional network that processes the input latent z and aligns it to the spatial grid of the pixel Transformer, ensuring that the latent preserves global structure and semantic information.
    • Sigma-aware gating is the key innovation that controls how much the model relies on the latent condition versus its own learned prior, with the latent influence adjusted by the noise level sigma through a gating function.
    AbstractPDF
  10. 222RoboticsarXiv:2605.22283

    Spatial Memory for Out-of-Vision Manipulation in Vision-Language-Action

    • The geometric estimator network, VGGT, estimates the camera pose and rough scene geometry to establish a global 3D coordinate system.
    • The time needed to find the target object was reduced by about 40 to 59 percent.
    • The robot showed fewer viewpoint corrections, meaning it less often had to readjust after the head camera located the target, which suggests that spatial memory provided sufficiently reliable estimates for the robot to commit to grasping.
    AbstractPDF
  11. 221RoboticsarXiv:2605.22183

    Action with Visual Primitives

    • [Withdrawn paper] The latest arXiv version, v4, states that Weilong Guo has withdrawn this paper.
    • The official version history dates the withdrawal to 20 September 2026; the stated reason is to make further experimental improvements before publication.
    • Official reason: We wish to complete further experimental improvements before publishing this work.
    • Previously described methods and performance figures were pre-withdrawal claims and are not presented as currently valid research findings or grounds for recommending adoption.
    • Official withdrawal notice and version history: https://arxiv.org/abs/2605.22183 .
    AbstractPDF
  12. 220AI / GeneralarXiv:2605.22681

    Forecasting Scientific Progress with Artificial Intelligence

    • The knowledge gap is the observed performance improvement when a model can access relevant information that existed before the cutoff date, and it measures how much of a model's failure comes simply from not knowing already known facts.
    • The prediction gap is the remaining difference between performance and actual outcomes even when the model has full pre-cutoff knowledge, and it captures the intrinsic difficulty of predicting the unknown.
    • The models consistently reported high confidence in their predictions even when empirical accuracy was low, and this calibration error makes it difficult for human researchers to know when to trust AI predictions.
    AbstractPDF
  13. 219LLMs / NLParXiv:2605.22821

    Tokenisation via Convex Relaxations

    • For the integer-only approach, linear programming selects only tokens it is almost certain about, meaning tokens with values of 0.999 or higher. This often yields a vocabulary smaller than the allowed budget.
    • Proven optimality limit: now we know how much room remains to squeeze tokenization for compression. Since BPE is already within 1% of the limit at scale, researchers can focus more on other properties such as language morphology or robustness rather than compression.
    • Non-greedy alternative: ConvexTok builds a vocabulary by considering the entire dataset at once, which makes better use of the vocabulary budget.
    AbstractPDF
  14. 218AI / GeneralarXiv:2605.22166

    Adapting the Interface, Not the Model: Runtime Harness Adaptation for Deterministic LLM Agents

    • Action realization failure, or 33.3 percent, occurs when the model has the right intent but fails to express it in an executable format, such as by omitting arguments, using the wrong delimiter, or generating free-form text instead of structured tool calls.
    • Environment contract mismatch, or 23.2 percent, occurs when the model generates a syntactically valid action but violates the concrete rules of the tool or API, such as by accessing a non-existent database column or using an invalid flight code.
    • Trajectory degeneration, or 33.6 percent, occurs when the agent falls into a degenerate state, such as repeating the same invalid action, stalling without progress, or failing to recover from an error even after receiving feedback.
    AbstractPDF
  15. 217AI / GeneralarXiv:2605.22763

    Advancing Mathematics Research with AI-Driven Formal Proof Search

    • Lemma hallucinations: Sometimes the agent claimed that a false lemma was a well-known result in the literature. Because the agent did not attempt to formally prove that specific lemma, the error was only discovered late in the process.
    • Formalization as a filter: A formal proof assistant provides ground truth that makes AI-generated mathematics useful to researchers. It removes the need for humans to check basic logical errors.
    • AI for strategy discovery: AI can help mathematicians by proposing high-level plans or identifying candidate counterexamples that humans might miss.
    AbstractPDF
  16. 216Bioinformatics / GenomicsbioRxiv:2025.07.24.666556

    Scalable, Fast and Accurate Differential Gene Expression Testing from Millions of Cells of Multiple Patients

    • Core problem: cells in single-cell RNA-seq are statistically dependent, which breaks the assumptions of traditional differential expression tests designed for independent bulk samples.
    • Bayesian framework: it restores the independence structure at the patient level, cleanly separating inter-individual variation from true transcriptional differences.
    • Multi-GPU variational inference enables computational scalability to more than 10 million cells across thousands of samples from clinical cohorts, atlas projects, and drug-response screens.
    • Practical impact: it makes single-cell differential expression analysis possible at scale for the first time, without the computational bottlenecks or mathematical limitations of existing methods.
    AbstractPDF
  17. 215AI Safety / TrustarXiv:2605.01428

    Hallucinations Undermine Trust; Metacognition is a Way Forward

    • The central argument is that most factuality gains come from expanding the boundary of knowledge, not from improving awareness of that boundary, meaning the distinction between what the model knows and does not know.
    • The model may fundamentally lack the ability to separate truth from error, which creates an inherent tradeoff between removing hallucinations and preserving usefulness.
    • The paper proposes faithful uncertainty, meaning that the model's linguistic uncertainty should match its intrinsic uncertainty, which offers a third path beyond either answering or refusing.
    • For agent systems, uncertainty awareness becomes a control layer that governs when to explore, what to trust, and how to weight retrieved information.
    • Empirical evidence, including poor generalization of truth probes, confident hallucinations, and failures of alignment methods, supports the existence of the discriminative gap.
    Full textAbstractPDF
  18. 214AI / GeneralarXiv:2605.22794

    MOSS: Self-Evolution through Source-Level Rewriting in Autonomous Agent Systems

    • A rejection gate is implemented in a pre-call hook to block inefficient batching patterns.
    • Root-cause remediation is possible because, by targeting the harness itself, MOSS can fix architectural flaws that text-based evolution cannot touch.
    • Production usefulness comes from the integration of failure batching, test verification, and safe container replacement, which makes the idea of self-rewriting practical in real applications.
    AbstractPDF
  19. 213Machine LearningarXiv:2605.21768

    Memory-R2: Fair Credit Assignment for Long-Horizon Memory-Augmented LLM Agents

    • Surprisingly, these performance gains were achieved by training on only two full dialogue trajectories, which demonstrates the effectiveness of the RL paradigm.
    • It shows strong performance on out-of-distribution benchmarks such as LongMemEval and MSC-Self-Instruct.
    • The training gains observed on the 3B model transferred to the 7B model as well.
    AbstractPDF
  20. 212Computer VisionarXiv:2605.21605

    GenEvolve: Self-Evolving Image Generation Agents via Tool-Orchestrated Visual Experience Distillation

    • Failure memory is a lesson about how to avoid failure.
    • The prompt pool divides 20,000 requests into knowledge-anchored categories, such as facts and events, and quality-anchored categories, such as spatial composition and text layout.
    • Teacher trajectories are created by using a strong model to generate initial successful trajectories and then having a VLM judge inspect them for correctness.
    AbstractPDF
  22. 210AI / GeneralarXiv:2605.22791

    Gated DeltaNet-2: Decoupling Erase and Write in Linear Attention

    • On multi-key retrieval tasks, where several similar associations exist, Gated DeltaNet-2 showed a substantial advantage, confirming that separated gates help the model distinguish competing pieces of information within a fixed-size state.
    • By selectively erasing specific channels, the model avoids overwriting or blurring existing stored data that may still be relevant when new information arrives.
    • SWA handles local, fine-grained interactions within a small fixed window, such as the most recent 2,000 tokens.
    AbstractPDF
  23. 209Machine LearningarXiv:2512.04695

    TRINITY: An Evolved LLM Coordinator

    • The core idea is that a compact coordinator, a 0.6B LLM plus a 10,000-parameter lightweight head, assigns Thinker, Worker, and Verifier roles on each turn and delegates work across multiple LLMs.
    • Optimized with a separable covariance matrix adaptation evolution strategy (CMA-ES), it outperforms RL, imitation learning, and random search under high dimensionality and strict budget constraints.
    • It achieves state of the art on standard benchmarks, including 86.2% on LiveCodeBench, and consistently outperforms individual models and existing multi-agent methods across coding, math, reasoning, and domain knowledge.
    • The coordinator's hidden-state representation provides rich contextualization, and block-epsilon separability in the optimization landscape enables efficient evolutionary search.
    • It generalizes robustly to out-of-distribution tasks without task-specific tuning.
    AbstractPDF
  25. 207Computer VisionarXiv:2605.18746

    ESI-Bench: Towards Embodied Spatial Intelligence that Closes the Perception-Action Loop

    • Directional bias: AI models often show confirmation bias, repeatedly choosing actions that confirm their initial impression rather than exploring new information, even though that initial impression may be wrong, and they often commit to an answer early in the process with high confidence regardless of whether the evidence is sufficient.
    • Intelligent action planning: Develop an agent that can plan specialized trajectories to resolve spatial uncertainty.
    • Uncertainty management and calibration: Build a model that can understand when it does not have enough information and adjust its confidence accordingly.
    AbstractPDF
  26. 206Machine LearningarXiv:2605.22817

    Vector Policy Optimization: Training for Diversity Improves Test-Time Search

    • Maze Navigation is a synthetic task in which an agent must collect items such as gold and diamonds while avoiding lava hazards. Because the goals are deliberately conflicting, for example, gold is often near lava, it is a perfect testbed for validating Pareto-front coverage.
    • MuSiQue is a multi-hop QA task that rewards both final answer accuracy and citation accuracy for intermediate reasoning steps.
    • ToolRL is a function-calling benchmark in which the model must satisfy structural constraints while also achieving high attribute-level accuracy.
    AbstractPDF
  27. 205Computer VisionarXiv:2605.22718

    WorldKV: Efficient World Memory with World Retrieval and Compression

    • Token pruning: tokens with high similarity scores are treated as redundant and discarded, and only the most distinctive tokens, the bottom P percent by similarity, are kept.
    • LPIPS measures perceptual similarity, and lower is better.
    • PSNR and SSIM measure signal and structural similarity, and higher is better.
    AbstractPDF
  28. 204Computer VisionarXiv:2605.22818

    MotiMotion: Motion-Controlled Video Generation with Visual Reasoning

    • Semantic consistency measures how well a video matches the user's high-level intent.
    • Accessibility is improved by reducing the need for manual frame-by-frame animation or precise technical specifications, which makes advanced video creation available to a broader range of users.
    • Educational benchmarking gives the research community a specialized tool, MotiBench, that focuses on the logical and physical aspects of video generation rather than on visual fidelity alone.
    AbstractPDF
  31. 201Computer VisionarXiv:2605.20743

    Draw2Think: Harnessing Geometry Reasoning through Constraint Engine Interaction

    • Construction fidelity is a model-level property that measures whether a VLM's intended construction matches the requirements of the problem.
    • Explicit reading channel: observations showed that 74 to 91 percent of the system's derived answers included at least one value explicitly returned by the engine, confirming that the VLM actively uses the engine as a source of truth rather than relying only on its internal prior knowledge.
    • Correction and replanning capability: after receiving an error message and using the object deletion tool, the model can correct its path. When a construction path leads to a contradiction or an invalid state, the model can backtrack and try an alternative strategy, just like a human mathematician using compass and straightedge.
    AbstractPDF
  32. 200Computer VisionarXiv:2605.18324

    Improved Baselines with Representation Autoencoders

    • This paper identifies three insights from systematically examining several design choices to simplify and improve RAE.
    • First, it studies a generalized formulation in which the representation is defined as the sum of the last k encoder layers, rather than only the final encoder layer.
    AbstractPDF
  33. 199Machine LearningarXiv:2605.21488

    Equilibrium Reasoners: Learning Attractors Enables Scalable Reasoning

    • Maze-Unique: the authors found that task ambiguity is poison for attractor learning, because datasets with multiple valid solutions, such as different shortest paths in a maze, make it hard for the model to form stable attractors. They created a maze variant with a unique path and showed that EqR scales to solving complex 30x30 mazes with high confidence.
    • Scale generalization: at test time, simply increasing the number of iterations lets the model generalize well beyond the complexity seen during training.
    • Self-diagnosis: the fixed-point residual provides an intrinsic confidence score that does not require an external verifier.
    AbstractPDF
  34. 198Computer VisionarXiv:2605.18740

    Vision-OPD: Learning to See Fine Details for Multimodal LLMs via On-Policy Self-Distillation

    • Question generation: For each small region, the MLLM generates a question q that can be answered only by seeing the details inside that region.
    • Grounding to the full image: The region's bounding box is overlaid on the full image I to form the student input x. A spatial instruction, such as 'look at the red box', is added to the question so the student knows where to focus.
    • Crop generation: The region is cropped and resized, often by 2x, to form the teacher input x'.
    AbstractPDF
  35. 197Computer VisionarXiv:2605.22819

    Cambrian-P: Pose-Grounded Video Understanding

    • Annotation: use the off-the-shelf pose estimator ViPE to generate external and internal camera parameters for the remaining clips.
    • Model size: as the LLM size increases from 0.5B to 7B parameters, both VQA performance and pose accuracy improve consistently.
    • Data size: increasing the amount of training data widens the performance gap between the pose-grounding model and the baseline.
    AbstractPDF
  36. 196Computer VisionarXiv:2605.18678

    Lance: Unified Multimodal Modeling by Multi-Task Synergy

    • For understanding, generation helps: training on generative tasks such as video editing makes the model develop a deeper understanding of temporal dynamics and spatial relationships, which improves performance on video understanding benchmarks.
    • For video generation, Lance scores 85.11 on VBench, the highest among unified models at the time of publication, with strong performance on object grounding and temporal consistency.
    • For multimodal understanding, Lance scores 62.0 on MVBench, which evaluates temporal cognition and video reasoning. This is an 11.3% improvement over the next-best unified model, Show-o2 (7B).
    AbstractPDF
  38. 194Machine LearningarXiv:2605.19269

    CODA: Rewriting Transformer Blocks as GEMM-Epilogue Programs

    • It maintains local state for operations such as the online log-sum-exp used in cross-entropy loss.
    • It recovers a substantial portion of GPU time that would otherwise be lost to data movement.
    • It provides a structured DSL that simplifies writing high-performance kernels.
    AbstractPDF
  39. 193LLMs / NLParXiv:2605.18747

    Code as Agent Harness

    • For semantic verification, code can easily be checked through tests for syntax and basic correctness, but it is still hard to verify whether the code actually satisfies the user's higher-level intent.
    • Transactional state management ensures that multi-agent systems can handle complex, concurrent edits to a shared codebase without causing merge conflicts or inconsistent state.
    • Human supervision is about developing an interface that lets humans inspect and intervene on the agents' internal code outputs without breaking the automation loop.
    AbstractPDF
  40. 192Machine LearningarXiv:2605.19407

    A Bitter Lesson for Data Filtering

    • Use a fixed number of epochs and train for four epochs.
    • Simplify the pipeline. Instead of spending thousands of GPU hours and a lot of human engineering effort on complex classifiers and heuristic rules, simply increasing training time on raw parsed web data can yield better returns.
    • Even trash has value. Robustness to mixed and random data suggests that, if the model is large enough, almost any data containing some statistical signal related to the target distribution can be useful.
    AbstractPDF
  41. 191AI / GeneralarXiv:2605.19943

    Probabilistic Tiny Recursive Model

    • Although powerful, deterministic recursion can converge to suboptimal solutions without any escape mechanism.
    • A common workaround combines test-time task-specific input perturbation with answer aggregation through voting.
    AbstractPDF
  42. 190AI / GeneralarXiv:2605.19376

    Generative Recursive Reasoning

    • Majority voting selects the most common answer among a set of samples.
    • Stochasticity shows that the probabilistic nature of latent states is a powerful tool for structured reasoning and overcomes determinism by preventing models from getting stuck in suboptimal deterministic trajectories.
    • Width-based scaling introduces a way to expand reasoning performance by increasing the number of parallel samples, or width, instead of simply increasing sequence length or model size.
    AbstractPDF
  43. 189AI Research AgentsarXiv:2605.20025

    AutoResearchClaw

    • It uses multi-agent debate for better hypothesis formation and result analysis.
    • It supports self-healing execution that can pivot or improve when experiments fail.
    • It provides verifiable reporting to avoid fabricated numbers and hallucinated citations.
    • It offers human-in-the-loop control with seven intervention modes.
    • It includes inter-run memory that turns past mistakes into future safeguards.
    AbstractPDF
  44. 188LLMs / Pretraining / EfficiencyarXiv:2605.06546

    Efficient Pre-Training with Token Superposition

    • Pretraining large language models is often extremely expensive and inefficient at scale, and it usually requires complex and invasive modifications to achieve high data throughput.
    • The paper presents Token Superposition Training, or TST, a simple drop-in method that greatly improves data throughput per FLOP during pretraining without changing parallelism, optimizers, tokenizers, data, or model architecture.
    • TST has two stages: a highly efficient superposition stage that combines many consecutive tokens into one bundle and trains with a multi-hot cross-entropy, or MCE, objective, followed by a recovery stage that returns to standard training.
    • The authors evaluate TST extensively at the 270M and 600M parameter scales and validate it on 3B and 10B A1B mixture-of-experts models, which demonstrates strong robustness across settings.
    • Ultimately, TST consistently outperforms baseline loss and downstream evaluation and reduces the total pretraining time of the 10B A1B scale by up to 2.5 times at matched loss.
    AbstractPDF
  45. 187Agents / Memory / Multi-AgentarXiv:2605.08060

    The Memory Curse: How Expanded Recall Erodes Cooperative Intent in LLM Agents

    • The problem is to reconcile the classical game-theoretic prediction that longer memory promotes cooperation with behavioral psychology findings that excessive human memory can undermine cooperation.
    • There has been no systematic large-scale study of how changes in interaction-history length affect cooperation in LLM multi-agent social dilemmas.
    • Prior LLM agent studies often used short interaction horizons, which may have missed the long-term effects of accumulated history on cooperative intent.
    • We performed extensive experiments over 500 rounds in four repeated social dilemma games with seven LLMs, systematically varying the accessible memory length from 0 to 80 rounds.
    • We analyzed more than 378,000 chain-of-thought reasoning traces with lexical analysis to quantify strategic intent, from future-oriented thinking to paranoia.
    • We performed targeted fine-tuning, memory purification, and asymmetric-memory experiments to causally identify the mechanism and develop mitigation strategies.
    AbstractPDF
  46. 186Agents / Self-Improvement / ReasoningarXiv:2605.13821

    Harnessing Agentic Evolution

    • Existing agentic evolution methods, whether procedural or agent-based, struggle with long-horizon search and often get stuck in local optima or suffer from context drift.
    • Procedural evolution is rigid because it relies on fixed rules that do not adapt when the characteristics of the search space change, which limits continuous improvement.
    • There is no stable and systematic mechanism for organizing accumulated evidence, such as candidates, feedback, and failures, and then using it to modify the evolution mechanism itself.
    • The paper formalizes agentic evolution as an interactive environment in which the accumulated history of all candidates, feedback, traces, and costs forms an observable process-level state.
    • It introduces AEVO, a controlled meta-editing framework in which an external meta-agent observes this process-level state and generates 'meta-actions'.
    • These meta-actions do not directly generate new candidates; instead, they strategically edit the mechanisms that control how future evolution proceeds, such as procedural rules or the agent operating context.
    AbstractPDF
  47. 185Agents / Science / MatharXiv:2605.06651

    AI co-mathematician: Accelerating mathematicians with agentic AI

    • Mathematical research is a complex, iterative process that requires broad exploration, intuition building, and continuous revision beyond formal proof.
    • Existing AI tools for mathematics are specialized, such as automated theorem provers or discovery engines, but they lack an integrated, stateful system that can coordinate these capabilities for long-term collaborative workflows.
    • Mathematicians currently have to manually integrate multiple AI tools and computation resources across different aspects of their work, because there is no comprehensive, interactive AI assistant.
    • AI co-mathematician is an interactive workbench that uses a hierarchy of AI agents built on the Gemini language model to manage and execute mathematical research tasks.
    • The system uses a project coordinator agent to refine user intent and delegates tasks to a parallel-workstream coordinator agent that uses specialized subagents for literature review, coding, proof, and computational exploration.
    • The system focuses on producing 'living working papers' with internal links, margin notes for citation, and iterative AI-led review loops, while supporting asynchronous interaction and flexible user steering.
    AbstractPDF
  48. 184Reasoning / Test-Time ScalingarXiv:2605.08083

    LLMs Improving LLMs: Agentic Discovery for Test-Time Scaling

    • Test-time scaling strategies for large language models are usually developed through manual design and heuristic tuning.
    • This reliance on human intuition limits the search over complex computation-allocation policies and makes them vulnerable to human cognitive bias.
    • Manual methods often produce strategies that are specific to a task or model, which leads to suboptimal accuracy-cost trade-offs across broader settings.
    • AutoTTS formulates test-time scaling as a controller synthesis problem, where a searcher LLM repeatedly proposes and refines policies defined in code.
    • The method uses an offline replay environment to collect reasoning trajectories in advance, which allows candidate controllers to be evaluated cheaply and deterministically without live LLM calls.
    • The framework introduces a beta parameterization to simplify the hyperparameter search space and provides fine-grained execution traces as feedback for robust discovery.
    AbstractPDF
  49. 183Agents / Multi-Agent / SurveyarXiv:2605.14892

    Beyond Individual Intelligence: Surveying Collaboration, Failure Attribution, and Self-Evolution in LLM-based Multi-Agent Systems

    • Individual LLM agents run into limits on sustained coordination, adaptation, and multi-step tasks across roles and environments.
    • Existing research on LLM-based agents is fragmented, with individual capability, multi-agent collaboration, and self-evolution treated as separate topics, which blocks a holistic understanding of their interdependence.
    • In tightly coupled multi-agent systems, error propagation and cascading failures are hard to diagnose automatically, and current systems lack mechanisms that turn diagnostic insight into structured self-improvement.
    • The paper presents the LIFE progression framework, which structures the operational lifecycle into four causally linked stages: Lay for capability-based build-up, Integrate for agent integration through collaboration, Find for defect discovery through cause attribution, and Evolve for autonomous self-improvement.
    • It systematically reviews and formalizes each stage, including taxonomies of individual agent components, multi-agent collaboration mechanisms, failure attribution methods, and self-evolution mechanisms.
    • It proposes and analyzes a closed-loop self-improvement framework in which failure diagnosis feeds directly into structural adaptation, such as collaboration reconfiguration or agent behavior improvement.
    AbstractPDF
  50. 182LLMs / Memory / ArchitecturearXiv:2605.12357

    $δ$-mem: Efficient Online Memory for Large Language Models

    • Large language models, or LLMs, struggle to use long interaction histories efficiently because of the quadratic compute cost of expanded context windows and context degradation.
    • Current memory mechanisms, including text-based, external-channel, and parametric approaches, suffer from information loss, retrieval noise, integration overhead, or the inability to adapt dynamically.
    • There is a need for a way to provide LLMs with continuous and adaptable memory without full model fine-tuning or large architectural changes.
    • delta-mem compresses continuous past information into a fixed-size associative memory online state, or OSAM, which enables efficient and dynamic storage.
    • The method operates through a sequential read-adjust-write cycle, where signals read from OSAM generate low-rank corrections to the frozen backbone's attention mechanism and a gated delta rule updates OSAM.
    • The system provides multiple write units, namely Token-State, Sequence-State, and Multi-State Write, so that memory updates can be adapted to different contexts and backbone capacities.
    AbstractPDF