Leaderboards, eval harnesses and ablations — the contested business of deciding which model is actually better.
Computer-use agents are increasingly evaluated in realistic desktop environments, but existing benchmarks provide limited coverage of professional engineering workflows whose outputs are persistent, s...
Vision-language models excel at 2D image understanding but remain limited in 3D spatial reasoning. Progress is hindered by limitations in current benchmarks. First, 3D datasets often rely on point clo...
Progress in inclusive household surveys has strengthened socioeconomic evidence for forcibly displaced populations, providing indispensable benchmarks on living conditions and welfare. However, these...
Large language model coding agents have recently become useful for software tasks, but weaker or open-weight agents still struggle to reliably interpret user intent and execute complex multi-step work...
Multimodal image registration is a key component of many clinical workflows, yet it remains challenging because corresponding anatomical structures often exhibit substantially different image intensit...
X-ray coronary angiography is the clinical gold standard for coronary artery disease during real-time cardiac interventions, but provides only 2D projections of inherently 3D vessels. Existing learnin...
Orthrus is a hybrid autoregressive-diffusion architecture that accelerates autoregressive language-model inference by generating multiple tokens in parallel while using a frozen autoregressive backbon...
We study relative position encoding for multi-view vision Transformers under camera heterogeneity, including varying fields of view (FoVs) or projection models. Existing rotary relative position encod...
The paper introduces Neuro‑JEPA, a sparse multimodal foundation model that learns unified representations of brain MRI across T1w, T2w, and FLAIR sequences using a latent predictive objective and a Mixture‑of‑Experts architecture. It was pretrained on over 1.5 million scans from 428,647 studies and systematically evaluates architectural, masking, objective, and sparsity choices for robust multimodal representation learning. Across 47 tasks from three health systems and 12 public datasets, Neuro‑JEPA consistently outperforms a simple CNN baseline, demonstrating its effectiveness for both clinical and research applications.
arXiv:2609.12544v1 Announce Type: new Abstract: Clinical de-identification relies on accurately identifying personally identifiable information (PII). However, manually annotated datasets are costly...
The paper presents NAROCE, a Neural Algorithmic Reasoning framework for online complex event detection (CED). It decouples rule learning from sensor semantics by pretraining a Mamba-based rule reasoner on synthetic atomic event traces and then adapting it to raw sensor inputs with limited labeled data. Experiments on a simulator‑generated benchmark show that NAROCE matches or surpasses strong baselines while using far fewer labeled sequences and computational resources.
AsyncFlow is an asynchronous streaming reinforcement learning framework designed to improve the post‑training phase of large language models. It introduces a distributed data storage and transfer module that enables panoramic data management and fine‑grained scheduling, allowing automated pipeline overlapping and dynamic load balancing. The framework also employs an asynchronous producer‑consumer workflow to reduce computational idleness by deferring parameter updates within staleness thresholds, and it is architecturally decoupled from training and inference engines, providing modular, customizable user interfaces. Experiments show an average throughput improvement of 1.59× over the state‑of‑the‑art baseline.
FINESSE is an agent‑based simulation framework that generates synthetic, structured datasets of multiple interdependent financial event streams, such as transactions, payments, account status changes, and policy interventions. Each stream has its own action space, schema, and variable types, and the streams are coupled through agents’ evolving latent states, allowing temporally rich interactions. The accompanying FINESSE‑Bench dataset supports four tasks—balance forecasting, transaction fraud detection, missed payment prediction, and next event prediction—and baseline results are provided using various time‑series and event‑sequence methods.
The paper investigates how large language models (LLMs) used as judges in absolute scoring tasks exhibit systematic biases that compromise reliability. It shows that a judge’s task accuracy strongly predicts both its judging accuracy and its directional bias, yet more capable examinee models consistently receive more lenient judgments. To mitigate these biases, the authors propose a calibrated weighted majority voting (WMV) ensemble that estimates judges’ error rates from inter-judge agreement patterns, achieving near-oracle performance without labeled data and improving both accuracy and fairness.
CRFCAN is a complex‑valued residual FFT convolutional attention network that jointly estimates channel and phase noise in sub‑THz OFDM systems. It embeds FFT and inverse FFT modules within residual groups to enable iterative feature interaction across time and frequency domains, and includes dedicated residual blocks for complex feature extraction and multiplicative phase‑distortion modeling. Simulation results show that CRFCAN outperforms conventional algorithms and state‑of‑the‑art deep learning models in NMSE and BER, while offering single‑shot, fixed‑complexity inference and good generalization to unseen phase‑noise models.
arXiv:2609. 12591v1 Announce Type: new Abstract: Foundation models are increasingly adapted through fine-tuning, model editing, and alignment procedures while retaining previously acquired capabilities.
SIMS: Scale-Invariant Merit-Function-Based Scalarization for Multi-Task Learning proposes a new scalarization method for multi-task learning that is invariant to the relative scales of task losses. By using a logarithmic transformation, SIMS converts the multi-objective problem into a single objective that preserves weak Pareto optimality and allows a smooth surrogate with controllable approximation error. Experiments on standard multi-task benchmarks show that SIMS consistently outperforms existing scalarization methods and achieves state‑of‑the‑art performance.
SIFPBPNet is a dual‑path neural network designed for cuffless blood‑pressure estimation from photoplethysmography (PPG) signals. It separates steady‑state and instantaneous features through a Steady‑state Feature Path (SFP) that uses a Graph Attention Network to learn individual‑specific long‑term patterns, and an Instantaneous Feature Path (IFP) that captures short‑term dynamics and fuses them with the steady‑state prior via cross‑attention. On a large wearable dataset, the model achieves mean absolute errors of 8.57 mmHg for systolic and 5.97 mmHg for diastolic BP, outperforming existing methods and showing that the SFP module can be added to other backbones for 2.8–13.1 % MAE reduction.
FEAT is a foundation model designed for extremely large structured data that replaces quadratic self‑attention with a linear‑complexity dual‑axis encoding architecture. It combines an adaptive‑fusion bidirectional state‑space model with convolutional gated linear attention to achieve permutation‑invariant representation learning in O(N) time. Experiments on 12 real‑world database benchmarks show that FEAT outperforms existing structured data foundation models on zero‑shot tasks and can be up to 50× faster in inference latency.
The paper introduces a feedforward graph architecture that uses several frozen large language models as computational nodes connected through a shared continuous latent space via learned linear projections. By jointly optimizing projection matrices through backpropagation, the system combines the representations of three small frozen models with two larger ones, culminating in a lightweight cross‑attention output node. With only 17.6 M trainable parameters, the architecture attains state‑of‑the‑art results on ARC‑Challenge, OpenBookQA, and MMLU, surpassing both individual constituent models and parameter‑matched learned classifiers.