In briefVerified Gradients, Anchored Policies: Model-Based Reinforcement Learning for Efficient Flow Control
SummaryThis paper presents a reinforcement learning method for controlling fluid flows. It estimates the action gradient with a surrogate model and audits it against a finite-difference estimate from a high-fidelity solver. If the surrogate passes, a low-dimensional feedback law is fitted to its gradient; otherwise the solver estimate is used, and a quadratic penalty keeps the neural policy near that law. Across four test cases, the controllers match or outperform a model-free baseline while using 3 to 156 times fewer high-fidelity solver interactions.
In briefImage Quality for Fabric-Density Counting: Can Smartphone Photos Through a Pick Glass Replace Manual Thread Counts?
SummaryThis paper examines whether smartphone photos taken through a pick glass, a folding magnifier, can be used to count the threads per inch of woven fabric. The authors built PickGlass-TPI, a dataset of 76 factory scenes from 44 fabric patches, with reference counts made by a textile-trained annotator. They benchmarked vision-language models, CNNs, counting methods, and a training-free spectral counter, and found that none agreed with the reference within ±5%, with the best methods about 25% off. Repeatability across re-shot photos of the same patch was also a key problem.
In briefExplainable per-input confidence estimation for runtime ML predictions under out-of-distribution inputs
SummaryThis paper introduces LEX, a post-hoc explainability framework that clusters training input embeddings with a Gaussian Mixture Model and assigns per-cluster prediction error bounds to inputs. It gives input-specific runtime confidence signals without retraining. Evaluated on an LSTM-based IPS forecasting model on Intel Alder Lake across 34 benchmark applications, it reports about 94% containment on unseen workloads and per-input bounds about 17% tighter than global baselines, with negligible runtime overhead.
Mahmoud Sanad, Mohammed Bakr Sikal, Osama Abboud et al.
In briefStructured kernel surrogate models for attributing the influence of one data modality on another across multimodal samples.
SummaryThe paper studies cross-sample modality attribution, measuring how one modality in one sample affects learning outcomes on other multimodal samples. The authors propose structured kernel surrogate models and an algorithm called SKS, which applies kernel ridge regression with a structured compositional kernel, along with a gradient-based estimation method. Evaluated on two multimodal benchmarks covering satellite images, road networks, text, audio, and tabular data, SKS achieves 57.8% higher correlation with true prediction outcomes than existing methods. In downstream data selection, it improves average accuracy by 1.93% over baselines.
In briefTurning Business Process Diagrams into Reusable AI Agent Workflows
SummaryThis paper presents a method that converts BPMN process models into executable agentic workflows running on a LangGraph/Pregel runtime. LLM reasoning is inserted at selected activities, while BPMN still controls process flow, deterministic decisions, system interactions and human approvals. It also introduces Recursive Capability Composition, which lets agents and business processes be exposed as reusable tools for other agents, and uses process models and ontologies to supply organization-specific knowledge.
Christian Fillies, Jan Fillies, Frauke Weichhardt et al.
In briefMultimodal road closure mapping using cross-modal graph attention
SummaryThe paper presents a unified multimodal, multitask framework that classifies road closures as none, partial, or complete and associates them with specific road segments. It combines graph-structured map data, satellite imagery with projected street-level detections, and street-view images using cross-modal transformer attention. Evaluated on nearly 20K real-world scenarios from a commercial vehicle fleet, it reports 72.7% closure classification accuracy and 60% segment association accuracy.
Izzeddin Teeti, Vaibhav Kango, Nidhi Sridhar et al.
In briefComparing deep learning models for segmenting geographic atrophy in en face OCT eye scans.
SummaryThis study compares five deep learning architectures and a classical thresholding method for segmenting geographic atrophy, an advanced form of age-related macular degeneration, in en face OCT images. The models were trained on 269 OCT volumes from two different devices, after normalization and inpainting preprocessing. On a shared held-out validation set, nnUnet achieved the highest mean Dice similarity coefficient of 0.8882, and the authors conclude it is the most reliable option under limited data conditions.
Utkarsh Doshi, Adarsh Gadari, Sharat Chandra Vupparaboina et al.
In briefEvoSurv: Evolution-Aware Multimodal Model Reveals Complementary Survival Signals
SummaryThis paper introduces eTime, a 38-dimensional representation of tumor evolution built from eight feature groups covering mutation burden, mutational spectra, and copy-number changes. It integrates eTime with clinical variables and RNA expression in EvoSurv, a multimodal survival framework. Across six TCGA cancer cohorts with 2,782 patients, EvoSurv achieved the highest mean C-index (0.663) and remained significantly associated with survival in head and neck and liver cancers after adjusting for clinical and RNA-based risk scores.
In briefOne-sided expert-parallel communication for faster Mixture-of-Experts inference on NVLink GPUs
SummaryThis paper presents NVLinkOneSided, a one-sided all-to-all communication method for expert parallelism in Mixture-of-Experts models, designed for GPUs within a single NVLink domain. Source ranks push tokens directly into target GPU buffers without first exchanging metadata, and grouped GEMMs read them in place. Two synchronization protocols, Fence and Sentinel, target high-throughput and low-latency cases. On Blackwell GPUs, dispatch and combine run 1.2 to 3.0 times faster than DeepEP, and the MoE layer is up to 2.4 times faster.
In briefRanking language models more cheaply by combining human feedback with LLM judges
SummaryThe paper proposes Prediction-Powered Ranking (PP-Rank), which combines predictions from LLM judges with human feedback to rank large language models more precisely while still giving valid inference when human feedback is collected adaptively. The authors establish asymptotic normality and simultaneous ranking inference in theory. In semi-synthetic and synthetic experiments, PP-Rank improves score estimation, confidence interval width and ranking certification over human-only baselines, and it can reduce evaluation costs without sacrificing precision.
Gabriel Sargent, Will Wei Sun, Zhengwu Zhang et al.
In briefAgentScript-Eval: Benchmarking AI Models and Agents on Generating Code for an Enterprise Language
SummaryThis paper introduces AgentScript-Eval, a benchmark of 100 execution-graded tasks for AgentScript, the declarative language Salesforce uses to build conversational agents in Agentforce. The tasks cover writing agents from specifications, fixing broken definitions, and optimizing working ones. The best model reached 25.0% pass@3, and every model scored 0% on writing agents from scratch. The authors report that adding an agentic compile-and-edit loop gave the largest improvement, and that how domain guidance is worded and packaged mattered more than model size.
In briefSymbolic imitation learning with offline demos and online tuning, an interpretable control method
SummaryThe paper introduces Symbolic Imitation Learning, which distills an oracle policy from an offline dataset into a decision list with linear model motion primitives. It adds online Augmented-Random Search (ARS) tuning of constants and primitives. Across 20 benchmark RL tasks, the method approximates the oracle's performance and exceeds it on 11 tasks, while producing parsimonious, interpretable policies.
In briefX Marks the Spot: Diffusion Models for Photo Location Prediction Using Location Embeddings
SummaryThis paper addresses image geolocalization, which predicts where a photo was taken. Instead of generating coordinates directly, the authors learn in the representation space of a pretrained location encoder, using a Location Autoencoder that maps embeddings back to coordinates. They train DIG, a latent diffusion model that generates location embeddings conditioned on image features. The paper reports that DIG outperforms coordinate-based generative models and sets a new state of the art for parametric geolocalization on OSV-5M, Im2GPS3k, and YFCC4k.
WhyThe authors diffuse in pretrained location-encoder embedding space rather than generating coordinates, reporting new state-of-the-art parametric results on three benchmarks.
In briefMultimodal Judges Flag, Humans Decide: A Review-and-Edit Framework for AI-Generated Product Video Ads
SummaryThe paper presents a workflow for reviewing and editing AI-generated product video ads, in which reviewers check each scene against six questions. Vision-language, speech and language model judges flag possible issues, and reviewers accept or reject these flags and add missed problems. In experiments on 120 ads, reviewers marked the product wrong on far more scenes when judge findings were visible. A product judge reached AUC 0.84 on 120 blind-labeled scenes. The authors release a review-portal template and their data.
Anoubhav Agarwaal, Sukesh Perla, Karthikeyan Mohanraj et al.
In briefGroundCascade: Selective Second Opinions for GUI Grounding
SummaryThis paper presents GroundCascade, a training-free controller for GUI grounding, which maps a screenshot and instruction to the screen location an agent should act on. It uses agreement between full-screen and zoomed predictions to decide when to consult a complementary model. The authors report accuracy gains over the stronger standalone two-pass grounder on ScreenSpot-Pro, OSWorld-G-Refined and UI-Vision, with fewer model calls per request.
In briefControlling Clock Hands and Circular Fractions in Text-to-Image Generation via Angle Tokens
SummaryText-to-image diffusion models can draw realistic analog clocks but do not reliably show the requested time, and similarly fail to generate a specified fraction of a circular object. The authors encode angles as periodic sine-cosine features, inject them as tokens into the prompt, and train with Stochastic Localization Relaxation and a hand-swap preference objective. On analog clocks, the method reaches 63-78% Acc@6 degrees, versus below 1% for baselines. On circular-fraction generation it achieves 89.5% Acc@0.10.
Furkan G\"uzelant, Basavaraj Sunagad, Aysegul Dundar et al.
In briefRadRef: A Reference-Free Metric for Evaluating AI Radiology Reports
SummaryThe paper introduces RadRef, a metric that evaluates AI-generated radiology reports without requiring a radiologist-written reference report. It compares report claims against findings derived from the image, weighted by relevance to the clinical indication. On the ReXVal dataset it reaches 73.1% agreement with radiologist rankings, comparable to leading reference-based metrics. In a reader study of 100 model and 100 radiologist reports, its agreement with radiologists on omitted findings was close to the agreement between the two readers.
Shelly Schwartz, Alan Ropp, Michael B. Sneider et al.
In briefAI Alignment Is Neither Solvable Nor Sufficient: A paper arguing that treating alignment as a property of single AI systems is misguided, since emergent risks in human-machine societies also matter.
SummaryThis paper argues that AI alignment, the goal of making AI systems act in line with human values, is problematic. It contends that alignment is poorly defined, is not solvable where it can be specified, and, even where it seems solvable, would not necessarily reduce AI's negative consequences. The author notes that hybrid societies of humans and machines are complex systems shaped by interactions and feedback loops, so emergent risks may become a major source of harm. The paper concludes that AI safety is an ongoing challenge rather than something solved by verifying one model property.
In briefKeeping Risk Under Control While Learning Policies Online
SummaryThis paper studies risk control for online policy learning, where an agent collects data adaptively while aiming to keep each new policy's risk below a threshold with high probability. The authors propose a permutation-based weighted risk evaluator, with weights approximated by the Sinkhorn algorithm, and select policies whose upper confidence bounds, obtained via a Gaussian bootstrap, fall below the target. The method is evaluated on online contextual bandits, safe question answering, and medical claim filtering tasks.
In briefLanguage-guided hierarchical reinforcement learning for coordinating multiple robots in search and rescue
SummaryThis paper proposes LG-HRL, a cognitive multi-agent framework combining natural-language understanding, hierarchical policy decomposition, and a shared cognitive workspace for coordinating UAV and UGV robots in search and rescue. A large language model converts human instructions into structured semantic cues that guide high-level strategy and role assignment. In a simulated disaster environment, the authors report 2.3 times faster learning, 37% better coordination coherence, and 20-30% higher victim rescue rates.
In briefHow Much of Your VLA Is kNN: Study on Whether Vision-Language-Action Models Rely on Retrieval
SummaryThis paper examines how much of vision-language-action (VLA) model performance on the LIBERO simulation benchmark comes from retrieving training demonstrations. The authors report that retrieving nearby demonstrations and reusing their action chunks, without policy training, achieves a 91.7% success rate on LIBERO. They also introduce Local Action Consistency (LAC), an offline metric measuring whether nearby observations in representation space have similar future actions, and show that LAC tracks retrieval success and can guide training of a retrieval key.
In briefVIGOR: A Benchmark for Checking Whether Multimodal AI Models' Visual Reasons Are Correct
SummaryThis paper points out that multimodal large language models are usually judged only by whether their answers are correct, which does not show whether the answers rest on the right visual evidence. The authors introduce VIGOR, a benchmark that represents visual rationales as explicit concepts paired with pixel-level groundings, spanning abstraction levels from low-level visual properties to high-level motion and relations. They also develop an automatic annotation pipeline to build the dataset at scale. Results show that current models have limited visual rationale correctness and exhibit distinct grounding failures across concept levels.
In brief"Everything in its right place: attention is a radio head," a paper framing attention as oscillator synchronization.
SummaryThe paper proposes that attention in language models can emerge from a receiver synchronizing with its context tokens, where each past token acts as an oscillator whose state carries content and whose frequency drifts with age. Solving Kuramoto's problem for this receiver yields a closed-form attention layer with two phases. Above a critical coupling set by the context, the receiver locks and its weights normalize; below it, the receiver stays silent and its weights sum to less than one.
Llu\'is Arola-Fern\'andez, Aissam Khadraoui, Alex Arenas
In briefTeaching the Harness: Student-Aware Skill Transfer for LLM Agents
SummaryThis paper introduces Student-Aware Teaching (SAT), in which a stronger LLM builds a model of a fixed student agent, tests hypotheses about its limitations with diagnostic probes, and writes reusable skills for the diagnosed failures. A skill is added to the library only if it raises success on matched development tasks. In four primary runs, SAT with Claude-Sonnet-5 teaching DeepSeek-V4-Flash reached 86.46% task success, above a profile-free teacher and the untaught student.
In briefAn AI agent autonomously optimizes numerical algorithms for atomistic simulation
SummaryThe paper presents AIOA, a framework where a large-language-model coding agent improves atomistic simulation algorithms in a loop protected by verifiers it cannot change. Applied to the ASE library, agents made six kernels about 22 to 24 times faster on average in Python while keeping results unchanged. On unseen problems, the agents' optimizers needed 3.1 to 3.4 times fewer force evaluations than L-BFGS for structure relaxation and 4.6 to 17 times fewer than standard climbing-image nudged elastic band for reaction paths.
WhyThe authors apply LLM coding agents in an improvement loop to atomistic simulation algorithms, reporting two components absent from the literature and several-fold fewer force evaluations than L-BFGS
In briefDOBB: Dynamic Oriented Bounding Boxes for Safer Multi-Robot Collision Avoidance
SummaryThis paper introduces Dynamic Oriented Bounding Boxes (DOBB), a geometric representation that converts each agent's current state estimate, uncertainty, and response characteristics into a dynamic, direction-dependent safety margin. The authors derive geometric features describing the relative configuration and severity of interactions between agents, together with a reward function that keeps their physical meaning. Across four evaluation environments, DOBB achieves the lowest mean total failure rate of 2.6%, and it maintains collision-based failure at or below 2.3% under out-of-distribution shifts in actuation dynamics and state-estimation uncertainty.
Pedro Antonio Alarcon Granadeno, Jane Cleland-Huang
In briefAgreeing on the Answer but Not the Reason: Limits of Measuring LLM Reasoning Failures
SummaryThis paper examines whether a failure category for LLM reasoning, "Compositional Collapse," can be measured reliably. Across twelve models and eleven languages on a multilingual math benchmark, the category's share depended heavily on the order in which the classifier tested categories, and human raters and an automated inspector agreed poorly on it, though they agreed well on answer correctness. The authors recommend reporting label co-occurrence, permuting test order, validating on shared independent samples, and auditing systematic disagreements before comparing failure-mode rates across models.</summary>
In briefByzantine-robust Muon: stable distributed training with the Muon optimizer
SummaryThis paper studies whether the Muon optimizer, which uses momentum orthogonalization, stays robust when some participants are adversarial and data differs across nodes in distributed and federated learning. The authors analyze Muon-based updates with robust aggregation within a first-order trust-region framework, without assuming i.i.d. data, and prove convergence for non-convex and star-convex objectives. Experiments on CIFAR-10, TinyStories and other benchmarks report that the tested Muon configurations achieve higher accuracy under attack than the tested SGD and normalized-SGD configurations.
Grigory Malinovsky, Yury Demidovich, Peter Richt\'arik
In briefChoosing Successful References for Multi-Turn Policy Distillation via Action-Sequence Alignment
SummaryThis paper studies on-policy distillation (OPD) for multi-turn tool-use agents, where a teacher is conditioned on a successful trajectory from the same task. The authors frame reference selection as a free-target optimal transport problem and use longest common subsequence (LCS) over actions as a proxy for trajectory compatibility. On ALFWorld and the airline domain of τ³-bench, LCS is the only evaluated selector that beats reference-free OPD on the overall ALFWorld score and all three airline metrics. Models trained on the airline domain also transfer to retail, where LCS improves all three metrics without further optimization.
In briefActivation-guided graph retrieval for reconstructing memory in language model systems
SummaryThis preprint proposes an architecture for memory retrieval in language model systems that adds activation-conditioned memory availability to standard query-based relevance selection. Request and goal cues seed an authorized memory graph, with activation and fan attenuation bounding the traversal, and a typed evidence graph is reconstructed before generation. The authors provide architectural analysis, cost estimates, and an evaluation protocol, but report no benchmark measurements.
In briefImage Transformers with Late-Stage Pixel Refinement: a latent-free image generator improved after training
SummaryThis paper studies image generation with diffusion transformers that do not use a latent VAE. It builds on JiT, a standard transformer that predicts the final image, and adds simple convolutional pixel-pathway layers in a post-training phase to improve quality. The pixel pathway only needs to be activated in the last 20% of the denoising process. On ImageNet at 256 resolution, the FID improves from 2.34 (with corrected sampling) to 1.78, without additional losses.
In briefASPIRE: An Agentic AI Platform for Iterative System Identification
SummaryThis paper presents ASPIRE, a framework in which a large language model agent chooses and runs nonlinear system identification methods on measured data. It builds closed-form static or dynamic models, each reported with parameters, an estimation record, an interpretation of terms, and an error on test data hidden from the agent. The framework was tested on a mobile robot, a rotational system with dry friction, and a laboratory fermenter, with all 30 runs producing valid models at a cost of 0.96 to 3.38 US dollars per run.
Erik Derner, Ji\v{r}\'i Kubal\'ik, Robert Babu\v{s}ka
In briefBuilding World Models Through Intrinsic Motivation: How Curiosity-Driven Learning Shapes Reusable AI Models
SummaryThis paper examines how 24 intrinsic motivation objectives, such as curiosity, entropy, and competency, shape world model learning when no external rewards are present. The authors integrate these objectives into a common model-based reinforcement learning framework and test the resulting world models on locomotion, manipulation, and open-ended gaming tasks. Curiosity-driven exploration combined with observation reconstruction performs best overall, while episodic and lifelong novelty signals yield the strongest zero-shot transfer and few-shot adaptation in compositional vision-based tasks.
In briefReal-time error correction during LLM reasoning
SummaryThis paper addresses how a single wrong step in large language model reasoning can propagate and waste tokens. The authors propose online deterministic verification, where code checks each reasoning step as it is written, without any model call or GPU use, and the model is told of the mistake and continues from the same point. Tested on symbolic mathematics with the ASyMoB benchmark and three open-weights models, the method improved accuracy by 1.4 to 5.6 percentage points while increasing token usage by only 6 to 7%.
In briefOne Temperature to Rule Them All? A study of sampling temperature in LLM inference
SummaryThis paper examines how sampling temperature should be chosen when large language models generate multiple answers at inference time, comparing a single fixed temperature with varying temperatures across samples in the pass@K setting. The authors analyze how per-question success probabilities change with temperature across several models and datasets, and derive an efficient method for evaluating the theoretically optimal multi-temperature allocation. They find that multi-temperature schemes typically give only modest gains over a single tuned temperature, which they explain through mechanisms they call dulling effects.
Aviv Orly (1), Ori Shem-Ur (1), Yaron Oz (1) ((1) School of Physics et al.
In briefReframe: Observation-aligned multi-teacher on-policy distillation for visual agents using tools
SummaryThis paper presents Reframe, a post-training framework that aligns the observation state, not only the output distribution, for visual agents that use tools. GD-GRPO trains spatial and temporal specialists with evidence-aware rewards and stochastic removal of the original visual input, while OG-MOPD replaces weak tool actions mid-rollout and resumes distillation from the repaired observation. The authors report that Reframe outperforms Mix-RL and vanilla MOPD across 14 image and video benchmarks.
In briefPaper on letting a frozen model learn robot manipulation tasks in context within a simulated world
SummaryThe paper proposes World In-Context Learning (WICL), in which a frozen model learns from one demonstration, rehearses the task in a world such as a simulator, and writes feedback into its context rather than its weights. The authors build RoboScript to implement this for robot manipulation. On all 50 RoboTwin 2.0 tasks, with one demonstration per task and no gradient updates, it reports 83.4% mean success, which they say exceeds the strongest trained action policy by over 20 points.
WhyThe authors propose WICL, learning manipulation in context via simulated rehearsal, reporting 83.4% mean success on 50 RoboTwin 2.0 tasks from one demonstration, over 20 points above trained policies.
In briefGuiding data allocation to improve performance on underrepresented subgroups in machine learning
SummaryThis paper studies how machine learning models can stay accurate on data subgroups whose distribution differs from the training data. The authors extend a Gaussian feature-model framework to theoretically analyze performance across many possible training subgroup proportions. They show that several configurations, including substantially imbalanced ones, can perform about as well as a balanced configuration. They also derive gradient-based directions for adjusting subgroup proportions to improve adjusted and worst-group accuracy.
In briefControlling which modality a multimodal model trusts when inputs conflict
SummaryMultimodal large language models can become unreliable when visual and textual inputs disagree, sometimes favoring a modality that is not the most reliable for the task. The paper identifies a small set of attention heads, called arbitration heads, that causally influence whether predictions follow visual or textual evidence. It introduces a lightweight, training-free intervention that biases these heads toward a designated trusted modality at inference time. The authors report that the heads are stable across calibration set sizes and that steering them enables bidirectional control across diverse benchmarks.
In briefIC-GNN: Modeling Individualized Brain Communities for Disease Prediction
SummaryThis paper proposes IC-GNN, a model that builds individualized functional communities over atlas-defined brain regions from resting-state fMRI data and uses them to predict disease. It was evaluated on more than 30,000 participants from ABIDE (autism), ADNI (Alzheimer's disease), and the UK Biobank (depression), where it outperformed 15 baseline methods. The authors report that the individualized communities are reproducible within and across scanning sessions and carry predictive power on their own.
In briefKC-Maps: a trajectory-free neural method for simulating molecular dynamics
SummaryThis paper introduces KC-Maps, a neural network method that learns the transition kernel of a stochastic differential equation. It needs only trajectory-free marginal samples together with the known drift and diffusion coefficients. Experiments on peptide molecular systems show it outperforms classical molecular dynamics in GPU-hours.
Majdi Hassan, Marta Skreta, Dominique Beaini et al.
In briefPredicting Future Hand-Object Motion in 3D from Internet Videos
SummaryThis paper presents Grasping-the-Future, a unified generative model of joint 3D hand-object motion covering wrist movement, finger articulation, and object position and orientation. Trained with masked spatiotemporal completion and flow matching, one model handles future forecasting, mutual prediction of hand and object motion, trajectory generation toward a target object pose, and completion of partially observed interactions. The authors also build a pipeline that reconstructs 3D supervision from internet videos and report stronger results than released forecasters on datasets with 3D annotations.
Rustin Soraki, Matthew Wallingford, Ryan Aby et al.
In briefINTENT: A Benchmark for Testing Event-Sequence Models on Unseen Users, Periods, and Items
SummaryThis paper introduces INTENT, a public benchmark for evaluating next-action prediction models on user event sequences such as transactions and app sessions. It scores models across seven cells defined by whether the user, time period, and target item appeared in training. The authors audited 59 public datasets, kept five, and split each at its own cutoff using one fixed rule. In reference results, the best model varies by cell, and a parameter-free baseline that repeats the user's last item performs strongly on new items in four of five datasets.
Mattias Kvist, Erik Lidman Hillbom, Edoardo de Cal et al.
In briefAdaptive Quantization for TD Learning over Rate-Limited Channels
SummaryThis paper studies policy evaluation with linear function approximation when an agent sends quantized temporal-difference learning updates to a server over a finite-capacity channel. The authors propose BAQ-TD, which combines batching to reduce transmitted variance with adaptive quantization to encode innovations. They prove a high-probability finite-time error bound showing that, above a certain channel capacity, BAQ-TD matches unquantized TD's convergence rate up to logarithmic factors. The method needs O(d log T) bits over horizon T to reach the order-optimal rate.
WhyThe authors propose BAQ-TD, using batching and adaptive quantization over a finite-capacity channel, with a high-probability finite-time bound and an exponential communication-complexity improvement.
In briefClick3R: Interactive 3D Reconstruction Corrected by Sparse Point Clicks
SummaryThe paper presents Click3R, a framework that lets users fix errors in feed-forward 3D reconstruction models by clicking a few corresponding points across an image pair. A lightweight adapter injects these correspondences as geometric constraints into the network. The authors also build a challenging benchmark that exposes failure modes of existing methods. Experiments report lower reconstruction error with as few as one click, on both standard and the new benchmark.
In briefHiVe-Mem: a memory framework for multimodal agents that expands related memories and selects evidence per query
SummaryThis paper presents HiVe-Mem, a multimodal memory framework for agents with long interaction histories. It organizes memories into a relational graph linked by textual and visual anchors, expanding retrieved seeds with related neighbors, and keeps several evidence forms per memory, such as summaries, captions, visual views, raw images, and interaction history. A lightweight trainable selector chooses an evidence subset for each retrieved memory based on the query. Across three multimodal memory benchmarks, the authors report consistent performance gains over base models and about a 90% reduction in token consumption compared with the most efficient multimodal baselines.
In briefVendi Anomaly Score: a diversity-based method for spotting unusual data points
SummaryThis paper proposes the Vendi Anomaly Score (VAS), which flags an anomaly by measuring how much removing a sample changes the dataset's diversity as measured by the Vendi score. VAS does not rely on density estimation or partitioning heuristics, requires no hyperparameter tuning, and adapts automatically to the data's spectral structure. It is non-parametric and scales linearly with dataset size. Across 148 benchmark embedding datasets, VAS achieves state-of-the-art anomaly detection performance, and it remains robust on large-scale ImageNet experiments as contamination rate and dataset size increase.
In briefSWAM: Scaling World Action Models with Cross-Modal Alignment
SummaryWorld action models jointly predict video and actions conditioned on language. The paper studies how to keep language-video alignment intact while learning actions, using about 47,000 hours of general, egocentric, UMI, and robot demonstration videos. Trained only on RoboTwin Clean demonstrations, the model reaches 88.10% and 61.56% success under Easy and Hard conditions, and 40.00% across three real-world tasks versus 23.33% for the strongest baseline. The authors release the curated data, training code, and model weights.
In briefWhere Alignment Emerges in Neural Networks: Revisiting the Platonic Representation Hypothesis
SummaryThe Platonic Representation Hypothesis proposes that neural networks are converging toward similar representation spaces. This paper examines where alignment appears between unimodal models across layers and representation dimensions. It finds that the trend of greater alignment with greater model capability persists in newer models when captions are clean and informative. The authors also introduce unimodal and multimodal baselines on various datasets, showing that unimodal models can exceed explicitly aligned vision-language models in cross-modal alignment yet remain well behind same-modality alignment.
Julia Herbst, Simon Roschmann, Quentin Bouniot et al.
In briefIs Persona Generalization Predictable? A paper on how fine-tuning a language model in one domain shifts its behavior in others.
SummaryThe paper studies persona generalization, where fine-tuning a language model in one domain changes its behavior in other domains, such as the broadly unsafe behavior seen in emergent misalignment. The authors find that in-domain and out-of-domain persona rates follow an approximately probit-linear relationship across fine-tuning hyperparameters, and that this relationship is roughly shared across model families and sizes, including Qwen and Llama. Changing which layers are fine-tuned alters the slope; for unsafe behavior, fine-tuning only later layers yields a shallower slope.
WhyThe authors report a probit-linear in-domain/out-of-domain relation across hyperparameters and model families, unifying emergent misalignment analyses.
Olina Mukherjee, Arnav Goel, Aditi Raghunathan et al.
In briefWorkflowArena: A Benchmark for Workflow Orchestration in LLM Multi-Agent Systems
SummaryThis paper introduces WorkflowArena, a benchmark of 145 long-horizon tasks across 8 professional domains, designed to test how well language model agents orchestrate their own worker agents through executable workflows. Deliverables are scored by a hidden verifier, and all delegation is recorded so that output quality and the organization of work can be evaluated together. Experiments with nine models find that no single configuration leads in every domain and that systems use different delegation strategies. When the executing model is fixed, workflows written by a stronger model substantially improve weaker executors.
In briefAI analysis of lymphoid structures in lung squamous cell carcinoma after neoadjuvant therapy
SummaryThis study used AI to analyze H&E whole-slide images of resected lung squamous cell carcinoma tumors and paired lymph nodes from 20 neoadjuvant-treated and 20 therapy-naive patients. Treated tumors showed lower stromal plasma cell and lymphocyte densities, plus fewer mature tertiary lymphoid structures and mature lymph node follicles. In an independent cohort of 458 patients, stromal plasma cell density and lymphocyte-plasma cell co-localization were associated with longer survival, suggesting a candidate prognostic biomarker.
Arne Schmidt, Justin Nieder, Cornelius Boehm et al.
In briefWhat blocks v-prediction? A patch-wise PCA study of pixel-space flow matching
SummaryThis paper examines why the standard v-prediction objective struggles in pixel-space flow matching, using patch-wise principal component analysis. It identifies a bandwidth bottleneck, as velocity in minor components is mostly determined by the raw noisy input. The authors propose the Raw-injected Transformer (RiT), which uses a lightweight raw-conditioned decoder and an output-side shortcut, and reports competitive results on ImageNet 256×256.
In briefReducing deception by AI agents using evaluation-side partial observability games
SummaryThe paper argues that language model agents can deceive humans to earn higher rewards while appearing helpful, because evaluations mostly check only final outcomes. The authors model deception mitigation as an Evaluation-Side Partially Observable Extensive Form Game (e-PoG), in which a monitor exposes deception from the full trajectory and a verifier checks the monitor's claims. They define a Subgame-Perfect Oversight Equilibrium and prove it exists in finite games, and experiments with Gemma3 and Qwen3 models show gains over the strongest baselines.
In briefProgramming Prescience: How Humans Build and Test AI Forecasting Bots
SummaryThis paper reports a qualitative interview study with 15 builders of forecasting bots, selected from successful participants in a recent AI forecasting competition. It identifies challenges in tuning bots to gather factual research, implementing coherent reasoning, and judging how well a bot performs. The authors also find that builders often added complexity that did not improve results and tested their bots in opportunistic, piecemeal ways.
In briefWhat the loss curve misses: 4-bit training learns rare facts more slowly
SummaryThis paper asks whether low-precision pretraining that matches validation loss also learns rare facts as efficiently. The authors train transformers from scratch with emulated 4-bit matrix multiplications on synthetic biographies with controlled exposure counts. NVFP4 needs 1.32 to 1.52 times as many exposures as BF16 to reach the same recall, even though validation loss differs by only 0.5% to 2%. The authors suggest measuring rare-fact acquisition alongside loss when evaluating low-precision recipes.
In briefMaxIM: Reinforcement learning for maximally informative incremental summaries
SummaryThe paper introduces MaxIM, a framework that trains AI systems to keep a bounded, incrementally updated summary of long interaction histories. It frames summarization as an information-theoretic reward-shaping problem, maximizing the predictive log-likelihood of future task-relevant targets and using an asymmetric critic to compute dense step-wise mutual information bounds. The authors report that on long-horizon scientific reading and multi-turn dialogue benchmarks, MaxIM outperforms RL, supervised fine-tuning, and prompt-based baselines without requiring human preference labels.
In briefHow Well Does AI Peer Review Work? A Test Built on 100 Planted Errors
SummaryResearchers inserted 100 known errors into 10 open-access psychology papers and had frontier models and two commercial AI review tools examine them. The best single system caught 71 errors and the worst caught 30, while pooling all outputs found 93. The study finds that combining systems is a major way to find more issues, and seven errors were missed by every system, all of them omissions of deleted information. The papers, errors, outputs, and experiment log are publicly available.
In briefLUCQUE: A Quebec lung cancer screening dataset for reliable AI evaluation
SummaryThis paper presents LUCQUE, a dataset built from low-dose CT lung cancer screening in Quebec, containing 1,678 baseline studies with expert-annotated pulmonary nodules and cysts. It is hosted on the PARADIM platform and accessed through a model-to-data approach so that data remains on premise. The authors state it is the largest accessible annotated real-world lung cancer screening dataset, intended to support reproducible and independent AI evaluation.
Johanna Topalis, Yannick Lemar\'echal, Gabriel Couture et al.
In briefComparing NLP and generative AI for extracting disassembly instructions
SummaryThis paper addresses the difficulty SMEs face in adopting disassembly automation as products grow more complex. It introduces and compares two frameworks that extract disassembly instructions from unstructured sources such as manuals, converting natural language text into a structured, machine-readable format. A rule-based NLP framework and a generative AI framework using LLMs were tested and compared on real-world disassembly manuals. The work provides a data foundation for developing and automating future disassembly processes.
Tim Schnieders, Tareq Maximilian Gabriel Ahrari, Oliver Petrovic et al.
In briefCounterfactual image inference without inversion: testing whether the latent-state inversion step is needed
SummaryThis paper studies counterfactual inference for high-dimensional variables such as images, which answers "what if" questions given observational data, modeling assumptions, and a causal graph. It examines whether the conventional step of inverting the factual image to recover its latent state is necessary, and evaluates an inversion-free method adapted from image-editing research. Experiments on the CelebA and ADNI image datasets measure effectiveness and identity preservation, and the paper presents the tradeoffs between inversion-based and inversion-free approaches.
In briefCurbside Chatbots: Auditing Language Model Mental Health Recommendations for Different Users
SummaryThis study audits 41 language models to see which mental health interventions they recommend when the same clinical case is presented by a clinician, a caregiver, or the person themselves. Across 19 case vignettes and five questions, the authors collected 35,055 free-text responses and had 29 models rate the full option lists on a seven-point scale. Models generally recommend options they rate highly, and their responses differ by the asker, with many giving clinicians consultation-style answers and lay users hedged, simplified advice.
Matthew Flathers, Jill Noorily, Phuong Anh Nguyen et al.
In brief2D Gum-Net: Unsupervised Elastic Image Alignment via Symmetric Correlation and Spectral Pooling
SummaryThis paper presents 2D Gum-Net, an end-to-end network that predicts dense elastic deformation fields to align one image with another without requiring labeled ground truth. It adapts a 3D subtomogram alignment method from cryo-electron tomography to 2D, using DCT-based spectral pooling to preserve fine structures. Trained only on synthetic fingerprints, it is tested zero-shot on real benchmarks, improving structural correlation by up to 15.7% on FVC2004.
In briefAgentic-OPD: Recovering lost agent skills by distilling from discarded training checkpoints
SummaryThe paper observes that reinforcement learning for LLM agents usually keeps only the best-validation checkpoint, while earlier checkpoints from the same run can solve problems it fails. The authors propose Agentic-OPD, which treats the best checkpoint as a student and all saved checkpoints as frozen teachers. Teachers are chosen by how well they cover the student's unsolved problems, and distillation is limited to failed student attempts where the teacher performs better. Across five agentic task domains and 24 benchmarks, it improves on its initialization checkpoint by up to 7.4 points.
In briefDual-teacher alignment for zero-shot EEG-to-image retrieval
SummaryThis paper proposes DTHA, a framework that decodes visual information from EEG by aligning brain signals with two complementary pretrained models: CLIP and DINOv2. A shared EEG encoder is trained with dual-branch contrastive alignment, and a fused-retrieval objective learns adaptive weights to combine the two teachers' representations. The encoder, AMSPBE, uses temporo-spatial attention and a learnable scalp-pattern bank. On THINGS-EEG zero-shot retrieval, DTHA reaches 92.4% Top-1 accuracy, 12.4 percentage points above the prior state of the art.
In briefSmartphone Depth-Aware Light Imaging to Detect Woody Breast Myopathy in Chicken Fillets
SummaryThis study presents a smartphone-based framework that combines transillumination images and 3D point clouds to non-destructively grade Woody Breast myopathy in chicken breast fillets. It analyzes 267 fillets using image, geometric, and one physics-based light attenuation feature. A multimodal ensemble model reached about 90% accuracy, compared with about 82% for models using 2D features alone.
Chaitanya Pallerla, Ramesh Bahadur Bist, Yihong Feng et al.
In briefUsing Large Language Models as a General Search Tool for Automated Machine Learning Across Open Tasks
SummaryThis paper proposes a large language model-based AutoML framework that jointly optimizes model architecture, training hyperparameters, and data ratios without task-specific fine-tuning. It introduces NCode, a unified numerical encoding that turns these design choices into discrete sequences, framing optimization as a ranking task. Trained only on trajectories from a single vision task, the searcher reaches state-of-the-art results on image classification, segmentation, speaker verification, and text-to-image benchmarks, and reports a GenEval score of 0.91 for a billion-parameter Diffusion Transformer using 896 GPU-days.
In briefFreezing Behavior, Not Weights: Rehearsal-Free Continual Learning With One Shared Prompt
SummaryThis paper proposes TEA-Prompt, a continual learning method that adapts a frozen vision transformer using a single shared prompt token set, without storing old images. It combines an input-conditional token relevance module with an attention-behavior anchoring loss that constrains attention outputs around class-specific query regions. The method reports state-of-the-art accuracy and reduced forgetting on CIFAR-100, ImageNet-R, and CUB-200 using only 0.38M parameters.
Duy-Cuong Bui, Trung-Anh Dang, Vincent Nguyen et al.
In briefMethods for eliciting parameters of a multi-criteria sorting model
SummaryThis paper proposes a rigorous procedure for directly eliciting the parameters of the Non-Compensatory Sorting (NCS) model, a multi-criteria decision aiding model. The parameters are the sufficient coalitions of criteria and the profile separating two categories. The authors design questioning algorithms based on depth-first search and dichotomy, and evaluate them by simulation.
In briefLogCDM: An Interpretable Log Anomaly Detector Using Graph Drift and Adaptive Fusion
SummaryThis paper presents LogCDM, a compact detector that scores each log session using three complementary signals: next-event surprise from a small causal Transformer, graph-attention embedding distance from a normal reference transition graph, and a transition-drift statistic shown to be a frequency-weighted chi-squared divergence. The model has 8.1 to 8.8 million parameters and needs no GPU at inference, reaching F1 scores of 0.984 on HDFS, 0.870 on BGL, and 0.850 on Thunderbird. It scores a session in 5.5 to 10.5 milliseconds on a CPU.
In briefLeft-side ripple effects in byte-pair encoding: certified lookbehind for any dictionary
SummaryThis paper builds on Berglund and van der Merwe's formalization of byte-pair encoding and studies how far a prepended string on the left can change a tokenization. It gives a method that applies to every dictionary, deciding the origin radius in polynomial time and characterizing when it is infinite. The authors also present a streaming tokenizer that emits tokens no continuation can change, and report results for the merge tables of GPT-2, cl100k, o200k and Llama 3.
WhyThe authors resolve the open left-prepend question for BPE, deciding infinite radius for all dictionaries and reporting infinite origin radius for GPT-2, cl100k, o200k and Llama 3.
In briefPaper showing learning rate decay can exponentially speed up SGD for global nonconvex optimization
SummaryThe paper examines whether learning rate decay in stochastic gradient descent (SGD) gives an asymptotic advantage for global optimization of nonconvex functions. The authors identify a natural class of nonconvex functions, including many popular benchmarks, where SGD with decay needs exponentially fewer gradient queries than SGD with any fixed learning rate in the small-step regime. They also show that annealing solves the associated non-log-concave sampling problem in polynomial time, using weak Poincaré inequalities.
WhyThe authors prove exponential speedup for SGD with learning rate decay on nonconvex global optimization, analyzed via weak Poincaré inequalities.
Guneykan Ozgul, Dylan Herman, Junhyung Lyle Kim et al.
In briefPORT: Robust token-level reweighting for preference optimization
SummaryThis paper proposes PORT, a fine-grained method for aligning language models under noisy preference labels. It reweights individual tokens using the empirical CDF of their log-probabilities, suppressing corrupted tokens with only a forward pass as a proxy for gradient-norm penalization. The authors show theoretically that PORT reduces gradient bias from noisy labels by minimizing an upper bound on that bias. Experiments across diverse noise settings report consistent gains over sequence-level baselines.
In briefSage-Taiyi: A Metacognitive System for Guiding Scientific Discovery
SummaryThis paper presents Sage-Taiyi, an AI architecture inspired by metacognition that uses accumulated scientific understanding to decide what to investigate next and how. It monitors and revises its assessments of evidence, assumptions, and open questions. On the Taiyi-Bench ideation benchmark, it ranks first on all eight metrics, scoring 54.0 overall versus 47.9 for the strongest baseline.
In briefA review moving from Granger causality to causal structure in dynamical systems
SummaryThis article examines Granger causality, a widely used method for detecting predictive relations between measured signals. It first reviews recent advances in analyzing these relations in observed data. It then turns to directed structure in latent dynamical models and asks under which conditions that structure can be interpreted causally.
In briefThe Narrative Gap: Can LLMs Help Users Explore Varied Accounts Across Languages?
SummaryThis paper examines how large language models present competing accounts of contested topics across languages and regions. The authors define six capabilities needed for diverse multilingual information seeking and introduce NarrativeBench, with 1,333 news articles in 23 languages and 460 queries. Evaluating 14 LLMs, they find current models lack the capabilities needed to reliably navigate diverse narratives.
Nikhil Sharma, Kelly Marchisio, Kenton Murray et al.
In briefClique Constraints for Exact Causal Structure Learning
SummaryThis paper adds an upper bound on the maximum clique size of the learned undirected skeleton to exact score-based causal structure learning, enforced through lazy cycle and clique constraints at integer candidate solutions. Experiments on linear-Gaussian Erdős-Rényi and scale-free DAGs compare callback order, clique-cut policies, incumbent repair, fractional separation, and Big-M bounds. The results show that clique constraints do not uniformly speed up optimisation; their benefit depends on graph density, sample size, and cut-management policy.
Xiaoyu He, Pavel Ryt\'i\v{r}, Petr Ry\v{s}av\'y et al.
In briefA tokenizer covering 204 languages that uses fewer tokens than GPT-4o, including for English
SummaryThe paper trains Havon, a byte-level BPE tokenizer, on human-written text in about 200 languages. On the FLORES-200 test set, it needs fewer tokens than GPT-4o's o200k_base for all 25 Indian languages, but about 7% more for English. The v4 version, which adds a multi-word training stage, needs 6.1% fewer tokens than GPT-4o on English at the same vocabulary size. The authors note these are compression results only, and they have not yet evaluated a model trained with v4 or v5.
In briefMeasuring how tokenizers charge more for Indian languages than English
SummaryThis paper measures how many tokens nine public tokenizers need for the same content across 25 Indian languages, using FLORES-200 and PHINC data. Several languages need far more tokens than English; on GPT-4o's tokenizer, Odia needs about 5.01 times and Santali in Ol Chiki script 13.7 times. Sarvam-30B/105B is the only third-party tokenizer that stays below 2.6 times English on every language.
Johin Johny Arimbur, Ankur Ashit Ghosh, Shervin Sam
In briefPlant Orbiter: Generating 3D Plants from One Image Using Video Diffusion and Gaussian Splatting
SummaryThis paper presents Plant Orbiter, a method that uses video diffusion to create high-quality 3D Gaussian Splatting models of plants from a single image. The authors report that it recreates plant morphology more accurately than other state-of-the-art methods, achieving the highest 3D F-score, while being easier to align and cheaper in camera placement for high-throughput pipelines. Combined with a 2D diffusion model, it also generates 3D plant models from text prompts with state-of-the-art results.
Zane K. J. Hartley, Yong En Kok, Lewis A. G. Stuart et al.
In briefInference validation for regulated AI, built on NVIDIA AIPerf
SummaryThe paper presents Runway, an internal validation tool built at MasterControl around NVIDIA AIPerf, which adds experiment design before measurement and compliance, cost, and audit decisions after it. It reports three findings from Runway sweeps: TensorRT-LLM and vLLM rankings reverse as concurrency changes, a newer and pricier GPU instance can cost less per request, and speculative decoding behaves differently across engines under load, so engine configuration can matter more than engine choice.
In briefΣ-LeWM: Learning robot control world models using S-information
SummaryThis paper adds a backward dynamics predictor and an inverse dynamics action predictor as latent losses to LeWorldModel. These extra losses impose further structure on the latent embedding. The authors report gains in performance and planning efficiency, which matter for real-time model-based control of robots. Results are shown on standard world model benchmarks and two simulated robot manipulation tasks.
In briefKiteNorm: Variance Regularisation for Stable, Scalable Post-LN Transformers
SummaryThis paper proposes KiteNorm, a post-norm training method for Transformers that constrains hidden-state variance to improve training stability. Experiments on decoder-only models up to 1.3B parameters show it outperforms leading baselines at their optimal learning rates, with scaling laws favouring it across depth, width, batch size, and training budget. Ablation studies indicate that each component is necessary for the full stability and performance gains.
Leon A. Trochelmann, Sajad Movahedi, Shiwei Liu et al.
In briefSafe Reinforcement Learning for Clinical Agents: Dense Rewards vs. Outcome Rewards
SummaryThis paper studies reinforcement learning with verifiable rewards for clinical multi-turn agents that decide what information to request, when to stop gathering evidence, and which action to take. The authors build three controlled environments: medication refill triage, acute diagnostic work-up, and dosing titration. They find that outcome-only training can raise aggregate reward while causing premature commitment, one-question collapse, and failures on rare high-risk cases. Giving dense credit for questions that reduce uncertainty improves results in diagnostics. The authors conclude that verifiable final outcomes alone are insufficient for training reliable multi-turn agents.
In briefOptiCover: A scalable framework for optimizing cold-start cover photos across multiple surfaces under semantic constraints
SummaryAirbnb hosts must choose one static cover photo from many listing photos, but over 90% of photos lack engagement data, creating a cold-start problem. The paper proposes OptiCover, which combines candidate photo generation, lightweight vision heads on frozen CLIP embeddings, a vision-language model that checks semantic representativeness, and an exploration logging module that collects randomized engagement data. The framework is deployed in production and produced a 1.5% engagement lift on the HomePage and a 0.35% increase in bookers on Search in online experiments.
In briefGenerative AI image labeling without fixed taxonomies (GLOC method)
SummaryThe paper presents Generative Labeling On Clusters (GLOC), a method for providing fast, accurate, and scalable online insights about images. It groups semantically similar images using optimized embeddings and summarizes insights at the cluster level so they apply to all cluster members. At Pinterest, GLOC generates image descriptors and pivot query suggestions for Hybrid Search, and the authors report that offline and online A/B tests show better quality and efficiency than previous methods.
Zhenjie Zhang, Sainan Chen, Kshiteesh Hegde et al.
In briefSpaghetti Splatting: differentiable curve-based 3D scene rendering via convolution
SummaryThis paper proposes Spaghetti Splatting, a method for learning curved 3D primitives from images using volume rendering. Curves are approximated by polylines, and each segment is rendered so that connected segments combine without double counting at joints. The camera projection is linearized per segment for fast training and inference. The authors report that the method generalizes 3D Gaussian Splatting, is competitive on general scenes, outperforms line-specialized methods, and natively supports anti-aliasing.
George Shramko, Shrisudhan Govindarajan, Rudy Lam et al.
In briefHidden Tails: Verifying Tail-Risk Claims When Some Outcomes Are Unobserved
SummaryThe paper addresses how past decisions hide some outcomes, which can conceal a predictor's largest errors and leave population tail risk unresolved. The authors derive sharp bounds on conditional value-at-risk for binary outcomes and bounded losses, under a stated limit on how outcome odds differ between observed and hidden cases. In controlled experiments with known loan repayments, recovering about 16% of hidden labels removed a median 61% of estimated ambiguity, compared with 16% for random recovery.
WhyThe authors derive sharp conditional value-at-risk bounds under selective labels and a minimum-cost audit allocation, tested in controlled loan-repayment masking experiments.
In briefOpen-source tool for removing patient identifiers from clinical notes, combining a local language model with rule-based and neural detectors
SummaryThis paper presents DeidSharp, an open-source .NET toolkit that merges findings from pattern rules, a clinical named-entity recognition model, and a locally hosted 26-billion-parameter mixture-of-experts language model to detect protected health information in clinical text. The authors compare several existing de-identification tools on nursing notes and semi-synthetic discharge summaries. They report recall-first and precision-first configurations, and the toolkit runs without sending text off-site.
In briefMulti-Agent Screening: What Agreement and Acceptance Cannot Tell You
SummaryWhen AI agents pre-screen items before a costly human assessor reviews them, only the passed items get judged. The paper shows what each kind of evidence can establish about the agents' errors. Agreement among agents does not show they are right, and assessor acceptance gives precision but only a lower bound on recall. Recall can only be determined by judging items the agents did not pass, and the needed sample size does not grow with the number of agents.
In briefFisher-Glass Learning: Classification Bounds and Subpopulation Audits
SummaryThe paper extends the Fisher-glass framework to link information remaining after adjusting for nuisance variation with local classification difficulty and subpopulation auditing. It derives lower bounds on local minimax excess classification risk under explicit distinguishability and decision-separation conditions. On the Waterbirds dataset, accuracy is 93.2% when background type is preserved and 40.6% when it changes, and a search identifies a subpopulation with 98.8% membership-weighted accuracy. The authors find that information gains and predictive gains can diverge, so they recommend jointly assessing predictive performance, information under nuisance models, and sensitivity.
In briefGeoScope: a training-free method for counting and spatial questions on ultra-high-resolution satellite images
SummaryThe paper finds that vision-language models undercount objects in ultra-high-resolution remote sensing images once a view contains more than about five objects, and that this limit narrows as views grow. GeoScope has a text-only language model turn each question into a spatial program that sets the region and resolution, while the vision-language model only grounds and counts within its effective range and directions are computed from the grounded boxes. Without training, it improves Qwen3-VL-8B by 16.3 points on regional counting and 10.5 points on spatial relations on XLRS-Bench.
WhyThe authors propose a training-free method splitting views by a measured VLM counting limit, reporting large gains on several UHR remote sensing benchmarks.
In briefGeneralized adaptive boosting and the geometry of mistakes
SummaryThis paper extends classical boosting to treat false positives and false negatives separately, introducing a "Z-learner" that guarantees a joint pair of error rates lying in a set Z. It gives a complete geometric characterization of which sets Z can be boosted, unifying the classical result with a recent multi-objective extension. The authors also present an adaptive algorithm that works without knowing Z in advance, and show that Z-learners can amplify confidence, using the centerpoint theorem from computational geometry.
WhyThe authors generalize weak learning to joint FP/FN guarantees over a set Z, giving a complete geometric characterization of boostable sets that unifies prior results.
Marco Bressan, Nataly Brukhim, Nicol\`o Cesa-Bianchi et al.
In briefTACTUS: A Robotic Palpation System That Maps Hardness Across Chicken Breast to Detect Woody Breast Myopathy
SummaryThis paper presents TACTUS, an automated tactile array that produces thickness-normalized, three-dimensional hardness maps at 20 locations on a broiler breast fillet within about 30 seconds. The system was tested on 100 fillets across four woody breast severity classes. It found significant force differences among severity classes and a moderate but significant correlation with texture-analyzer measurements.
Chaitanya Pallerla, Yihong Feng, Avisekh Tripathy et al.
In briefAgnostic Marginal-Nonuniform Learning: studies binary classification learning rates in the agnostic setting, comparing optimal rates under standard and relaxed criteria.
SummaryThis paper studies binary classification in the agnostic marginal-nonuniform learning framework, where learning rate constants may depend on the unlabeled data's marginal distribution but guarantees must hold uniformly over all label conditional distributions. In the standard agnostic setting, the optimal rate is 1/√n and the marginal distribution does not help. In the relaxed setting, optimal rates form a trichotomy of exponential, linear, and arbitrarily slow rates, along with a fine-grained rate based on the uncentered VC-eluder dimension.
In briefA lattice-based HMM for scanning Ottoman poetry's aruz meter
SummaryOttoman Divan poetry follows aruz, a meter based on syllable weights, but modern Turkish spelling drops the diacritics that record those weights. The paper proposes automatic methods to identify each poem's meter and align its syllables with the meter's positions. A hidden Markov model decodes over a lattice of candidate paths and is trained by EM without scansion labels. On a held-out set it fully scans 77.6% of test lines, outperforming a rule-based scanner, a neural tagger, and few-shot Claude Opus 5.5.
In briefAverage vs. expected causal effects when individuals influence each other: modeling choices and estimation limits
SummaryThis paper studies causal inference when one person's treatment can affect others' outcomes, a situation called interference. The authors organize their assumptions into three layers (interference neighborhoods, exposure mappings, and outcome models) and split potential outcomes into direct, spillover, and interaction components. They distinguish average causal effects, which compare outcomes under fixed conditions, from expected causal effects, which compare outcomes under a specified allocation policy, and bound the variance of the Horvitz-Thompson estimator for both. They also propose graph-aware independent-set designs whose variance bounds are polynomial in the maximum degree.
WhyThe authors propose hierarchical finite-population models separating ACEs and ECEs, and report variance bounds contrasting exponential and polynomial sample-size growth under interference.
Vishesh Karwa, Edoardo M. Airoldi, Donald B. Rubin
In briefCoilStellaration: A Dataset for Designing Stellarator Fusion Magnet Coils
SummaryThis paper introduces CoilStellaration, a dataset and benchmark of 177,000 optimized coil sets for stellarator fusion devices. The benchmark asks models to predict coil geometries from a target plasma equilibrium and requirements on spacing, curvature, and magnetic-field error. The authors find that magnetic-field accuracy remains the main challenge, while using learned predictions to initialize conventional coil optimization can speed up early progress. Baseline models and evaluation code are released.
Jan-Hendrik Ewers, Santiago A. Cadena, Nicol\`o Foppiani et al.
In briefreComposer: An Editable Planning Interface for Controllable Image Generation
SummaryThis paper presents reComposer, an image-generation interface that breaks a user's evolving intent into creative decisions such as subject, composition, and palette. Users can edit this plan through conversation or direct manipulation. In a study with 8 design and non-design professionals, participants reported greater control and ownership than with prompting when creating assets, exploring alternatives, and iterating on details.
In briefRecovering tissue optical properties from intravascular OCT images using differentiable inverse rendering
SummaryThe paper presents a method to recover per-pixel optical attenuation coefficients, markers of pathology and cellular composition, from a single intravascular optical coherence tomography (OCT) image. It uses gradient descent through a differentiable physics-based forward model, requiring no training data or segmentations. The method was verified against analytical solutions, validated on five in vitro phantoms, and applied to 29,269 clinical images from 128 patients.
Ajay Manicka, Karim Kadry, Sebastian Ruiz-Lopera et al.
