Video fusion aims to integrate complementary cues from heterogeneous sources, producing visually informative videos with coherent spatio-temporal structures. Jointly learning cross-modal complementarity and temporal dependencies typically relies on large-scale, high-quality video datasets, whose storage and repeated use during optimization incur substantial costs, making data-efficient learning highly desirable. To address this challenge, we rethink video fusion from the perspective of dataset condensation and propose STAR-VF, the first framework that condenses spatio-temporal representations for data-efficient video fusion. STAR-VF condenses the original training set into a compact set of learnable spatio-temporal representations, which are decoded into synthetic video clips and optimized using feature-level supervision provided by a frozen proxy model. The resulting condensed data preserves fusion-relevant spatial and temporal cues without requiring ground-truth fused videos as supervision. To adapt fusion learning to condensed representations, we equip the final fusion model with flow-free inter-frame temporal modeling and query-driven intra-frame aggregation, enabling direct temporal interaction and adaptive selection of fusion-relevant information in representation space. Extensive experiments on infrared-visible, multi-exposure, multi-focus, and medical video fusion benchmarks demonstrate that STAR-VF, trained with only approximately 1/16 of the original temporal training set, achieves fusion quality comparable to methods trained on the full data. Further analyses validate the contributions of learned condensation and flow-free temporal modeling, and demonstrate the robustness of the condensed supervision across different proxy encoders and downstream fusion models. Project page: https://zju-xyc.github.io/STAR-VF-Project-Page/.
Individuals experiencing suicidal ideation (SI) or psychological distress may disclose their social identities when seeking support from LLMs. Attending to identity-related stressors can be important for providing contextually appropriate and empathetic support. However, disclosing an identity does not necessarily mean that it is a cause of distress. In this work, we audit how LLMs interpret distress through disclosed identities. Using paired counterfactual interventions, we evaluate six LLMs across two SI and distress datasets, 16 identity and control conditions, and two prompting protocols. Our findings suggest that models can transform identity disclosures into explanatory narratives, attributing distress to presumed identity-related adversity. These assumptions frequently appear within otherwise empathetic responses, particularly when marginalized identities are disclosed. We contextualize our findings through the lens of compassionate othering to characterize these patterns as a form of harm in which empathetic responses impose identity-linked assumptions on individual suffering. These responses can misrepresent individuals' experiences by treating their identities as explanations for distress without supporting evidence. Our findings highlight the need to mitigate these harms by grounding interpretations of distress in individuals' disclosures and avoiding unsupported identity-linked assumptions.
Hierarchical Joint-Embedding Predictive Architectures (H-JEPA) enable long-horizon visual control but rely on costly test-time gradient descent for planning. Our analysis reveals that iterative optimization consumes up to 97.7% of episode runtime and that low predicted rollout costs do not necessarily translate into successful control. We further identify a planning-horizon asymmetry between long-horizon top-level composition and short two-step subordinate planning. To address these challenges, we propose WARP-JEPA, a hierarchical amortized planning framework that replaces online gradient optimization with learned multi-candidate trajectory proposers while keeping the world models frozen. WARP-JEPA combines a lightweight world-model-in-the-loop causal Transformer at the top level with parallel residual MLPs at subordinate levels. The proposers are trained using model-based rollout costs and best-of-K behavior cloning, enabling efficient candidate selection without test-time gradient optimization. Across 30 frozen H-JEPA models on four benchmarks, WARP-JEPA achieves 9.6-18.8x faster episode inference on long-horizon tasks (2.9-3.0x on Cube) and improves task success by up to 33.3 percentage points. These results demonstrate that amortized planning can substantially improve both inference efficiency and control performance without retraining the world models. Code is available at https://github.com/hideinsewer/WARP-JEPA.
Bilevel optimization addresses machine learning problems in which the outer objective is a value function implicitly defined through the solution of an inner problem. While its algorithmic aspects are well developed, its generalization properties remain much less understood. In this paper, we study the generalization of inexact gradient methods when both objectives are least-squares regression problems and when the estimator of the value function's gradient, or hypergradient, is obtained by implicit differentiation in function space. This approach allows us to derive stationarity bounds for these methods in terms of the uniform generalization errors of the two regression functions needed to estimate the hypergradient. When these regression functions are defined as overparameterized two-layer ReLU networks trained by gradient flow with early stopping, we obtain explicit finite-sample bounds on these generalization errors in the Neural Tangent Kernel (NTK) regime. Combining these with our general inexact gradient analysis shows the effect of sample size on the overall convergence of the bilevel method. Numerical experiments on synthetic data illustrate our results. Beyond the NTK regime, further experiments reveal an initialization-dependent implicit bias induced by the bilevel structure, which can lead to feature learning at the inner level and overall performance improvements.
Hybrid quantumclassical models are increasingly proposed for nancial prediction, but reported gains are rarely separated from the classical components around them. We study a hybrid quantum temporal convolutional network for daily cross-sectional ranking of 2000 Tokyo Stock Exchange equities under a protocol built to make the quantum contribution identiable: six conditions (circuit trained, frozen or absent, a parameter-matched classical surrogate, and the circuit alone, trained or frozen) through one code path; four walk-forward folds with ve seeds; bootstrap intervals, Deated Sharpe Ratios, backtest-overtting probabilities and family-wise error control; ve classical baselines including a gradient-boosted tree; and a locked holdout. On 500 out-of-sample days two one-sided tests place the trained-versus-frozen dierence inside 0.03 annualised Sharpe (p = 0.0055, minimum detectable eect 0.02): the circuit is equivalent to a frozen one, not merely indistinguishable from it. With the circuit as the head's only input the two are equivalent inside 0.25; against the network with no circuit the design resolves only 0.50. No condition is distinguishable from zero (best De-ated Sharpe Ratio 0.17). The holdout reproduces the ablation on a protable window, and 2000 US equities reproduce it where every condition loses money.
We improve Turkish mathematical reasoning in Large Language Models by combining LoRA fine-tuning, weight-space model merging, and search-based inference. Using Turkish-Llama 8B-DPO with Turkish translations of MetaMath, NuminaMath, and TigerLab, we find data quality beats scale (Mix-30k best among fine-tuned; large MetaMath underperforms). Linear merging delivers the largest gains (GSM8K up to 0.63), while Ties is modest and Breadcrumbs/Slerp degrade. Search with MCTS and Tree of Thoughts improves mean LLM-judge scores for reasoning quality on a 0-100 scale. Limitations include machine translation, limited scale, and search stochasticity.
Hardware designs may be available only as gate-level netlists or input-output descriptions when the original register-transfer-level (RTL) design and documentation are not accessible. Low-level representations obscure word-level computation and its finite-width behavior. Recovering high-level RTL from such representations therefore helps engineers understand a circuit's functionality and enables RTL rewriting and resynthesis. A plausible function model is also insufficient unless the generated RTL is equivalent to the original circuit for all input combinations. In this paper, we present RTLREX, a large language model (LLM)-guided framework that coordinates domain-specific tools for function recovery, RTL generation, independent verification, and cost-guided rewriting. Independent verification establishes equivalence between the recovered RTL and the original circuit. The LLM uses tool feedback to select further analyses, revise candidate RTL, and propose syntax-cost reductions. Experiments on 30 benchmarks from Problem A of the ICCAD 2022 CAD Contest show that RTLREX recovers equivalent high-level RTL for all benchmarks, surpassing the state-of-the-art method WolFEx in recovery coverage. On benchmarks recovered by both methods, RTLREX reduces mean syntax cost by 20.5% relative to WolFEx. On complete truth-table benchmarks from the IWLS 2026 Contest, our framework reduces mean syntax cost by 95.7% relative to an LLM agent baseline on the benchmarks validated by both methods.
This study investigates sequential multi-task learning using Parameter-Efficient Fine-Tuning (PEFT), LoRA, and 4-bit quantization on Llama 3.1 8B Instruct for mental health problem detection, emotion detection, and sentiment analysis. Using 4,243 English social media posts, it compares two task sequences and demonstrates that task ordering significantly affects knowledge transfer and task interference. The MHP-first sequence achieves the best mental health problem detection (Micro-F1: 0.7624), while the Emotion-first sequence performs better in emotion (F1: 0.6052) and sentiment analysis (F1: 0.9459). A Joint-MTL baseline further highlights the challenges of simultaneous learning, including task interference and generative format collapse. Overall, the findings demonstrate the effectiveness of sequential PEFT for mental health NLP while showing that task order is critical to balancing performance across related tasks. The integration of LoRA and 4-bit quantization also reduces computational and memory requirements, supporting the adaptation of large language models in resource-constrained environments.
Parameterized instantaneous quantum polynomial-time (IQP) circuits generate distributions conjectured to be hard to sample classically. As reinforcement learning (RL) policies, their $2^n$ measurement outcomes are typically mapped to a smaller action space by a classical decoder. This aggregation can mitigate initialization concentration while making the policy classically simulable. We characterize this tradeoff through the decoder's Fourier--Pauli expansion. Under independent uniform initialization, we express policy variance as an exact sum over groups of nonconstant Pauli components with identical active-generator signatures. Each group contributes its squared aggregate coefficient times $2^{-r}$, where $r$ is the binary rank of its active generators. An inverse-polynomial aggregate coefficient with logarithmic rank prevents exponential policy concentration. Sparse Pauli expansions and contiguous action bins permit efficient classical additive estimation of action probabilities. We introduce a majority decoder combining protected low-weight components with exponentially large exact Fourier support, preserving initialization nonconcentration while excluding polynomial-size enumeration of the exact expansion. Numerical experiments corroborate the variance laws. In ten-qubit CartPole experiments, majority achieves higher returns than recursive parity and the tested local readouts, while higher-weight Pauli sectors contribute an increasing relative share of gradient magnitude during training. A complementary 21-qubit block-chain policy outperforms an extensive pool of fully connected classical neural networks in episode efficiency, even though its bounded-width structure permits efficient classical simulation. These results guide decoder design for quantum and quantum-inspired RL, distinguishing policy nonconcentration, learning performance, and classical simulability.
Zeroth-order (ZO) optimization estimates descent directions from function values alone, which removes the activation storage required by backpropagation and makes it attractive for memory-efficient fine-tuning of large language models (LLMs). Existing low-rank ZO methods either parameterize the additive update through Euclidean factors with a non-unique representation, or restrict each finite-difference direction to be low-rank. In this work, we propose to constrain each additive update block to a fixed-rank matrix manifold while keeping the pretrained weights frozen, so the learned update remains rank-constrained throughout fine-tuning. On this geometry, we build a Riemannian ZO gradient estimator that projects a random matrix onto the tangent space, normalizes it to correct the scale mismatch induced by projection, and further propose Zeroth-order Riemannian Gradient Descent (ZO-RGD) and its smoothed variant, ZO-RGD-S. We show that the ambient finite difference matches its retracted counterpart, the normalized estimator recovers the Riemannian gradient in expectation, and ZO-RGD converges under trajectory-regularity conditions. Experiments on LLM fine-tuning tasks show that the proposed algorithms are competitive with strong forward-only baselines across model families and tasks, with gains in several settings.
Under strict privacy regulations in the clinical domain, de-identification is necessary to enable the use of clinical data such as for natural language processing. However, we currently lack sophisticated ways to tell which de-identification models actually are suitable. Specifically, we need benchmarks that are challenging, diverse, and clinically plausible, all while ensuring label correctness. To this end, we present SPARCLING, an LLM-based pipeline to generate synthetic discharge letters with character-exact PHI span annotations. Our pipeline combines structured attribute sampling, template generation, and red teaming to obtain diverse PHI-labeled discharge letters. In our experiments, we demonstrate this superior content diversity over existing public datasets. Further, we show SPARCLING's downstream utility as a benchmark to uncover failure modes of de-identifiers in the German clinical domain. Finally, our experiments show that training on our synthetic data can yield more capable de-identifiers than training on the scarce public data alone. Our code is available at https://github.com/Matt00n/phi_discharge_letter_generator.
Recursive self-improvement offers a promising path toward AI agents that improve through experience by refining their own harnesses without retraining their underlying large language models (LLMs). Making this self-improvement process reliable, however, requires more than generating new updates: agents must decide which changes are worth keeping. A greedy gate that accepts updates based solely on higher validation scores can mistake an observed gain for genuine progress and lead to performance degeneration, whereas overly conservative statistical gates may reject beneficial updates and hinder performance improvement. We introduce Trustworthy Recursive Updates with Statistical Testing (TRUST), a framework that adapts an online multiple hypothesis testing procedure to control the rate of false evolution across rounds, with an LLM analyzing historical context to help determine how much evidence is required to update the incumbent. Under interpretable assumptions, we prove that TRUST limits the proportion of accepted updates that result in unacceptable performance loss. We further derive a lower bound on the final performance gain. Across the benchmarks HotpotQA, DROP, and APPS with the models DeepSeek-V4-Flash and Qwen-3.8-Flash, TRUST reduces the average ratio of accepted inferior updates from the greedy gate's 45.59% to 2.78%, while achieving higher final scores than the state-of-the-art statistical gate baselines for recursive self-improvement. These findings demonstrate that TRUST strikes a balance between statistical reliability and performance gain in recursive self-improvement.
We introduce a proximal preference optimization framework that unifies the implicit reward formulation of Direct Preference Optimization (DPO) with the trust-region stability of Proximal Policy Optimization (PPO). Our framework yields two complementary algorithms. P$^2$O (Proximal Preference Optimization) introduces a clipped importance ratio directly into the DPO preference margin, producing a reward-model-free algorithm that supports safe dataset reuse, bounded policy divergence, and monotonic preference improvement guarantees. PKTO (Proximal Kahneman-Tversky Optimization) extends P$^2$O by replacing the DPO margin with KTO's prospect-theoretic utility margins, adding a self-regulating adaptive baseline that tightens the effective trust region as the policy drifts, and an asymmetric weighting that handles imbalanced or asymmetric preference datasets. We provide a unified theoretical analysis, including Kullback Leibler (KL) trust-region bounds, a policy-improvement lower bound, and convergence guarantees under multi-epoch proximal updates. Empirically, P$^2$O and PKTO match or exceed DPO and KTO in alignment quality while maintaining better stability under repeated dataset reuse across multiple LLMs on three preference benchmarks. The code base is available here.
Electroencephalography (EEG) is widely used to measure brain activity for brain-computer interface applications for its non-invasive nature and high temporal resolution. However, progress in modeling the neural responses to visual semantics has been limited by a reliance on fixed image collections that lack the fine-grained control needed to isolate fine-grained changes in visual features and their relation to perception. In this work, we present EEGFaceSem, a large-scale dataset pairing high-quality EEG with photorealistic face images synthesized from a generative model. We provide the generative latent vector for each stimulus, enabling research into the continuous mapping between a visual latent space and neural representations. Each participant's brain responses were recorded while they attended to specific semantic features (e.g., smiling, hair color) in a rapid serial visual presentation paradigm. The dataset contains EEG signals from 30 participants, totaling 67,200 time-locked EEG recordings. We present benchmarks for single-trial semantic classification under single-subject, cross-subject, and subject-adapted settings, as well as for cross-subject voting that aggregates per-stimulus predictions across participants. We further demonstrate the dataset's unique utility with two complementary applications: deriving semantic directions in the latent space for brain-supervised image editing, and conditioning a generative model on brain-decoded labels for brain-supervised image generation. Our dataset paves the way for advancing brain-computer interfaces (BCI) for semantic-level visual saliency detection and brain-supervised generative modeling. Our dataset and code are openly released at OSF (https://osf.io/2guk3/) and at HuggingFace https://huggingface.co/datasets/yefllower/EEGFaceSem.
Camera rigs with identical measurement budgets can have different velocity estimation rates when exposure timing is uncertain. We establish this distinction for rolling-shutter reconstruction with two calibrated moving anchors, unknown scene points, and independently unknown auxiliary poses and clocks. On the same fixed local physical class near a separated lateral rig, repeated readouts have worst-case squared velocity risk at least of order $\sigma\rho_{\mathrm{eff}}$, whereas complementary readouts have minimax risk of order $\sigma^2$, for a fixed admissible clock displacement $\rho_{\mathrm{eff}}>0$ as the noise standard deviation $\sigma$ tends to zero. A finite physical compensation path gives the lower bound; a nonlinear velocity inverse proves complementary stability. All motion, structure, pose, and clock nuisances remain unknown. We identify mounting configurations in which both designs become regular and quantify the risk floor induced by metric uncertainty. We quantify full-parameter domains with independent 6-ms clocks and test in-domain reconstruction under paired noise and a depth-offset intervention. Raster studies examine image formation and calibration sensitivity.
We prove universal approximation for the bilinear-readout equivariant message-passing network (BREMP), a reference architecture for equivariant graph neural network interatomic potentials. Its layers combine equivariant convolutions with affine maps, and its readout forms invariants from products of the equivariant features. We construct a one-layer feature that determines the local environment up to permutations of atoms of the same type. The construction uses spherical harmonic filters of degree no larger than the largest number of neighbors at a common distance from the center. Iterating the construction recovers the environment reachable in as many hops as there are layers, using the same bound on filter degrees at the layer cutoff. The readout approximates every continuous invariant function of the features uniformly on compact sets. Together these give universal approximation of atomic energies at a prescribed physical range using a shorter layer cutoff. We assume that, throughout the admissible class, message passing reaches all neighbors in each physical environment within a common number of layers. Finally, DPA4, an interatomic potential architecture in production use, inherits this result by approximating the BREMP layers constructed in the proof and realizing the bilinear readout.
Visual generative models increasingly blur the line between synthesizing observations and simulating environments. Yet visual plausibility alone does not constitute successful world modeling: compelling outputs may lack persistent state, physical dynamics, or causally consistent responses to interactions. We define a generative world model as a predictive dynamical system that maintains a persistent, history-conditioned representation of an environment, updates that representation in response to actions or interventions, and generates temporally coherent trajectories that reflect outcomes under observed and counterfactual interventions. From this definition, we derive an intelligence-oriented framework spanning perceptual grounding, persistent state modeling, controllable transformation, dynamics-aware prediction, and closed-loop interaction. We present WorldFoundry, open-source infrastructure for reproducible inference and evaluation across video, 3D and 4D, embodied action, and interactive world systems. WorldFoundry unifies model and benchmark discovery, runtime and asset preparation, execution orchestration, artifact management, and interactive visualization, while modality-specific adapters preserve native inputs, execution semantics, output representations, and evaluation protocols. Explicit manifests, durable artifact contracts, and evidence-aware scorecards separate execution readiness from evaluative claims, making each claim traceable to its supporting artifacts. The catalog contains 270 model entries, 74 benchmarks, and 62 registered metric entries, with readiness and execution evidence recorded separately. WorldFoundry enables reproducible comparison of persistent, controllable dynamics across heterogeneous systems, advancing the field from perceptual synthesis toward measurable world intelligence.
Pretrained encoders from self-supervised representation learning are increasingly applied to sonar imagery of the seafloor, where they are evaluated with protocols that take one embedding per target. A target patch of seabed, by contrast, is re-imaged from an alternate look on overlapping survey passes, so each embedding is one measurement of that target. Standard protocols cannot say how consistent those measurements are, or how much of their spread belongs to the patch and how much to the acquisition geometry. We treat this as a measurement-reliability problem. Using a synthetic-aperture sonar (SAS) survey, we measure the repeat-look reliability of embeddings from a frozen natural-image encoder (DINOv3 ViT-S/16). Standard single-look metrics, effective rank, clustering and a linear probe for coral, are all stable to the look, and the probe performs well (AUC 0.93). Yet the repeat-look reliability of the same embeddings is poor, with a distance-based intraclass correlation (dbICC) of 0.20 and nearest neighbours and clusters that barely agree across looks. This contradiction resolves as we find the probe learns a stable direction, whose ICC of 0.60 is 3.0 times that of the full embedding. Variance components, calculated from generalizability theory, attribute under a tenth of the within-cell spread to look acquisition geometry. The rest is cell-specific and cannot be separated without repeat passes at matched geometry. A sonar-pretrained encoder (BenthicDINO) and a classical SAS texture descriptor show the same pattern. For this dataset and the tested representations, uses that depend on patch identity fail, such as cross-pass comparison and retrieval, while uses at the level of habitat type (e.g. coral classification) do not. Because the standard metrics did not detect this, we argue that repeat-look reliability should accompany any evaluation of representations of the seafloor.
Search allows a robotic policy to evaluate candidate futures before committing to an action. Motivated by test-time planning for vision-language action (VLA) models, we study selective search over action chunks in PushT using a frozen Action Chunking with Transformers (ACT) policy. Beam search generates perturbed alternatives around the policy prediction and scores them through rollouts from cloned simulator states. Across 50 evaluation episodes, beam search raises task success from 0% to 18% and increases average cumulative reward from 39.44 to 103.2. We introduce One Step Ahead (OSA), a gate that simulates the nominal policy chunk and invokes full search only when its predicted coverage drop exceeds a threshold. In the reported ablations, OSA preserves or improves success while reducing runtime by approximately 3x relative to full beam search. Ablations over tree depth, branching factor, beam width, and replanning frequency show that larger search budgets do not consistently improve performance. The results support selective test-time search for action-chunking policies; evaluation on language-conditioned VLAs and physical robots remains future work.
Multimodal Large Language Models (MLLMs) have shown promise as interactive agents, yet precise, programmatic 3D graphics editing remains difficult to train because existing systems largely rely on closed-source model APIs and lack deterministic, state-verifiable execution environments. We present BlenderFORGE, a trainable data-environment-optimization framework for open-weight MLLM agents that edit explicit Blender scenes through Blender Python (bpy) code generation. BlenderFORGE targets four fundamental state-verifiable editing task families: object placement, shape-key editing, lighting adjustment, and material modification. It integrates three components: (1) a scalable perturbation-based data generation pipeline built from curated assets and large-scale 3D scene repositories; (2) an isolated Blender Sandbox for deterministic code execution, structured state export, and closed-loop reward computation; and (3) a two-stage training pipeline that uses offline teacher trajectories for cold-start Supervised Fine-Tuning (SFT) and sandbox-computed rewards for multi-turn tool-agentic Reinforcement Learning (AgentLoop-RL). Experiments across the four task families show that the resulting open-weight Qwen3-VL-8B and Qwen3.5-9B BlenderFORGE agents substantially improve over their base models and achieve competitive performance with several zero-shot proprietary MLLM baselines on selected metrics. Additional cross-benchmark evaluations and camera-reframing experiments provide evidence of transfer beyond the original training setting.