Diffusion-based vision-language-action (VLA) models cascade pretrained vision-language models (VLMs) with diffusion action experts, but typically restrict the VLM to providing static conditioning. This separation limits how much pretrained perception and reasoning can guide action generation, especially in out-of-distribution settings. We introduce JADE, a Joint Action DEnoiser that enables the VLM and action expert to jointly refine actions. Our key insight is that structured action latents allow the VLM to read the current noisy action together with visual observations and language instructions. At every denoising step, the VLM contextualizes the noisy action and guides the action expert through gated cross-attention, so that its guidance evolves with the action being generated. We further supervise the VLM to recover clean action latents from noisy ones, which gives it a direct action-learning signal and couples perception with control. Additionally, joint denoising enables JADE to generate actions with fewer sampling steps, helping offset the additional cost of VLM participation. An adaptive step-size schedule further balances task performance and inference latency. On RoboDojo, JADE achieves an average success rate of 12.67%, the highest among the evaluated $\pi_{0.5}$-based methods, and is competitive with recent VLA and WAM foundation models pretrained at scale. It also improves performance on RoboTwin 2.0 and raises the average success rate across four real-world bimanual tasks by 20.8 percentage points compared with $\pi_{0.5}$.
Open-source LLMs provide a flexible alternative to closed-source models for Text-to-SQL but still struggle with complex queries. Finetuning with chain-of-thought (CoT) data improves reasoning, yet standard datasets lack CoT annotations, requiring synthetic generation. Existing synthesis methods only verify the final answer, allowing flawed reasoning steps to corrupt the training data. We propose VCoT-SQL, a framework that improves Text-to-SQL generation through intermediate-step validated CoT supervision and a dedicated SQL validator, which enables small open-source models to outperform frontier closed-source models. VCoT-SQL trains two complementary models on curated reasoning traces: an SQL generator using clause-level CoT templates where reasoning steps are verified via execution and schema alignment checks, and an SQL validator using diagnostic CoT templates to detect syntactic, semantic, and type level errors. We further design a hierarchical schema filtering pipeline that integrates and refines complementary retrieval strategies to distill relevant database elements. VCoT-SQL achieves 66.82% execution accuracy on challenging dataset BIRD dev and 86.8% on Spider test, improving the base 7B model by 7% on BIRD. Notably, our CoT synthesis relies on 7B models yet outperforms methods that use 72B-scale or even proprietary closed source teachers for data generation, demonstrating that rigorous step-level validation can be more effective than scaling up teacher model size. Our dataset and code are available at https://github.com/INTREBID/VCoT-SQL.
We estimate the integral curve of an unknown vector field from noisy tangent vector field on a known compact Riemannian manifold. We consider two sampling settings: independent observations at random locations that have already been collected, and a fixed response budget that allows subsequent locations to be chosen using pilot information. Under suitable regularity assumptions, the passive kernel estimator has curve error of order $n^{-2/(d+3)}$ in probability, with sample size $n$ and manifold dimension $d$. Its functional central limit theorem has a limit characterized by a linear stochastic differential equation. For active sampling, an independent pilot directs one new batch of measurements to neighborhoods of the target curve. Under localization, smoothness and design conditions, this concentration yields expected maximum curve error of order $N^{-1/2}$, including the pilot in the total response budget $N$. Both methods provide simultaneous geodesic confidence tubes with asymptotically exact coverage under mild conditions. Simulations show narrower active tubes in most reported settings, with improved small-sample coverage in some three-dimensional examples. Ocean-drifter observations illustrate mean-flow reconstruction on the sphere, and a canopy-flow LDA application uses retrospective active sampling to quantify lateral and vertical departures of local mean streamlines.
Liquid-phase peptide synthesis (LPPS) offers a scalable route to peptide manufacturing, but route development still depends on empirical solvent screens because each protected intermediate must be soluble for reaction and recoverable for purification. The protecting groups, terminal modifications, and noncanonical residues that govern this behaviour are poorly represented in existing peptide property benchmarks, limiting the ability to develop and rigorously evaluate molecular representation models for LPPS-relevant prediction tasks. We introduce PeptSol, an open benchmark of 1388 hand-curated protected-peptide solubility measurements with chemistry-aware splits. The measurements span only 84 parent sequences and 101 protected structures, and 92.5% come from a single historical programme. Across seven frozen molecular encoders, the best macro-F1 is 0.843 on binary solubility for the historical subset under a peptide-held-out split; the harder three-class ordinal task reaches 0.614 in that cell. No foundation model consistently outperforms a baseline combining chemical fingerprints with explicit molecular descriptors, and model rankings vary across prediction tasks and data splits. The dataset, evaluation protocol, and baselines establish a standard for measuring progress on molecular representations for synthesis-aware peptide developability.
Does eliciting five weighted estimates within a single response consistently improve numerical judgments? We compare direct and multi-estimate elicitation across anchoring, hindsight, base-rate, conjunction, and status-quo/default tasks in six frozen configurations. Each configuration contributes 618 admitted development units and 3,702 admitted formal units; the six-configuration formal cohort contains 22,212 units, including 21 missing outcomes. Of 45 preregistered confirmatory tests across five configurations, 12 reject on complete cases and 11 retain directional confirmation after the prespecified missingness-protection rule. Only one adjusted bound exceeds its smallest effect size of interest: gpt-oss-120b shows a 10.68-percentage-point increase in post-outcome confirmation error under multi-estimate elicitation (worst-case estimate 10.04 points). The remaining ten retained effects have sub-threshold point estimates; non-rejection does not establish equivalence. Joint presentation lowers signed conjunction excess in four confirmatory configurations, but this comparison changes both constraint visibility and call structure. Flash-high combines two disjoint execution-window batches under an approved continuation protocol, preserving their identities and non-causal interpretation. Post-hoc candidate-spread associations across all six configurations remain in-sample and require independent predictive validation. The completed cohort provides no evidence for uniform debiasing by multi-estimate elicitation. Materials: https://github.com/asj252/llm-judgment-studies
Defense systems for maritime assets must distinguish approaching benign craft from uncrewed surface vessels (USVs) with sufficient warning to act. Shore cameras using computer vision can separate hostile vessels from benign traffic, but these systems may be attacked using an adversarial hull patch or physical disguise. We evaluate adversarial hull patches and find they suppress detection in up to 29% of attempts but never cause misclassification as benign. Physical disguise, however, can bypass both a camera check and a radar presence check. To evaluate defenses, we develop a hybrid replay protocol from automatic identification system (AIS) tracks that combines recorded benign motion with direct or off-axis attack run-ins. We find that for recreational-labeled vessels attacking via direct run-in within 4 nautical miles, an asset-relative closing-rate check flags 94-95% at 15-20 knots, compared with 12% for a track-only speed check. The checks are complementary on off-axis approaches, constraining the attacker to remain below both thresholds and extending the time required to reach the asset. Flagging on either check has a 4.9% held-out false-alarm rate overall and 12.3% on recreational traffic. The replay-based evaluation protocol supports comparative test and evaluation of defense methods.
Dimension reduction rests on a claim: t-SNE and UMAP preserve local structure, PCA global structure. Our DLP curve is the share of each point's k nearest neighbors a map keeps. Corrected for chance, it gives L at k=10 and G over 30 to 50% of the data, local and global fidelity. At the midpoint the curve is provably a rank correlation of distances. The LG reversal test decides, with a stated error rate, which of two maps is local and which global. On twelve datasets of up to 1.38 million points, G agrees with established global scores at 0.84 and 0.93, and the area under R_NX at 0.22 and 0.30. The entity fidelity E, read at each point's entity size, predicts which entities a map loses beyond L and G where entity sizes differ. In an agent memory of one million passages, t-SNE and PCA cross at 150,000 passages (p=2.6e-5).
Dense labels of the aorta in CT are slow to draw, and the public aortic CT datasets that could stand in for them each label different structures under their own conventions. Large language model (LLM) agents can now write code and read images. We asked whether such a model labels these scans better as a free-form agent that writes its own segmentation code, or inside a harness, where an automated pass labels the scan and the model reviews the result. We built a harness, Taurus, and compared it with free-form agents by blind label recovery: we withheld the human labels of six public datasets (939 volumes) and scored every approach against each dataset's own classes. In the first version of Taurus, the model's review raised the primary metric on only two of six datasets, by less than the noise between two reviews of the same case. A free-form agent running Claude Opus 5.5 scored higher on most datasets. Moving what the free-form agents did well into the automated pass then carried most of the gain on 38 held-out cases. The final version smooths the label surfaces with a fixed rule and checks for the large errors the automated pass tends to make, such as an aneurysm drawn to half its size. Each check that fires puts a multiple-choice question to the model, with candidate corrections drawn on the scan. It matched the surface smoothness of the human references, and in the primary metric we found no difference from the free-form Opus agent, at under half its cost, $1.22 against $2.71 per case. Most of what made Taurus competitive came from the harness and its automation rather than from the language model itself. Taurus shows that this kind of annotation can be automated, so that expert time can move from drawing labels to checking them.
Egocentric observations provide a robot-centric view of manipulation processes, capturing contact interactions and object-state changes that are important for visuomotor policy learning but difficult to observe from external cameras. However, collecting such data at scale remains expensive due to additional hardware requirements and platform-specific deployment costs. Synthesizing egocentric observations from large-scale exocentric manipulation videos provides a scalable alternative. However, existing methods struggle to synthesize physically consistent and policy-relevant egocentric videos under substantial viewpoint transformations, severe occlusions, complex object interactions, and inherently sparse observations from available exocentric viewpoints. Code will be made publicly available at https://github.com/xueha0/AuroraWorld
Floor plan generators are judged mainly by Compatibility, a graph edit distance to the requested adjacencies that is near saturation and blind to room shape. Our sample-level protocol calibrates each threshold no standard states at a declared false-rejection rate on real samples and grades them with the same checks. It reports each sample at its first failure, counts samples without a verdict apart, and attributes a stop to the generator or to its representation where a control exists. We apply it, with a production layout solver's checks, to four published generators against two built corpora, RPLAN and Modified Swiss Dwellings (MSD). A squareness check set per room type refuses 5 per cent of RPLAN apartments and 33.7 per cent of House-GAN++ plans. The gap survives resolution, tracing and orientation controls. Pooled over rooms, a Wasserstein distance on squareness orders the generators nearly as the check does but separates House-GAN++ from the built corpora by far less. Taken per plan on the check's worst-room margin, it separates them about fivefold to sevenfold. Within the GAN generators, Compatibility barely predicts which plans pass. We release the checker, which reproduces the solver's first failing check on 3,998 held-out RPLAN plans.
Terminal feedback in multi-turn agent training leaves both rollout allocation and local credit underdetermined. Dynamics-Predictive Sampling (DPS) [9] provides a task-level predictive sampling reference, but interactive agents also need to choose which visited states merit further continuations. We introduce TSBPO with two method components. A lightweight, fine-grained Root/Branch budget allocator distributes factual rollouts across tasks and additional suffixes across replayable states using likelihood-fitted estimates of terminal-outcome contrast. On the resulting physical rollout trees, a tree-aware step-level credit assignment algorithm compares immediate sibling segments and assigns their local outcome contrasts to the turns they contain. Exact task/observation groups provide a complementary step term. The allocator selects what to sample; observed outcomes determine the subsequent credit. With a Qwen2.5-1.5B actor, fixed checkpoints attain 97.79% seen and 96.64% unseen stochastic success on ALFWorld, and an 81.07% per-episode success rate over six stochastic episodes for each of the 500 WebShop test goals. At the WebShop update-100 checkpoint, removing the observation-group step term yields 75.93% under the same evaluation. These results describe the evaluated training configurations on two interactive benchmarks.
Collecting naturalistic human manipulation demonstrations with accurate contact force and hand pose is essential for robot dexterous manipulation policy and foundational world model learning - yet no existing wearable platform provides both calibrated fingertip force sensing and occlusion-resilient hand tracking in a form factor that preserves natural manipulation behavior. We present Tactus, a textile-integrated sensing glove that co-optimizes three design pillars: (1) a thin, conformal 3D-knit form factor (1.5 mm fingertip stackup) designed to minimize interference with natural manipulation, (2) strategically placed capacitive tactile sensors at manipulation-relevant regions informed by empirical contact studies, and (3) IMU-fused pose tracking that maintains accuracy through the hand-object occlusions where manipulation data is most valuable. Tactus includes a documented calibration pipeline for cross-glove consistency and per-session calibration for subject-dependent correction. To enable objective comparison, we introduce standardized benchmarking protocols for tactile force regression and hand pose estimation, and apply them to Tactus and two commercially available tactile gloves - to our knowledge the first quantitative cross-platform evaluation of this kind. Tactus resolves a minimum detectable force of 0.59 N at 95% confidence, 1.6 - 2.7x lower than the commercial baselines, and with 0.87 cross-participant consistency, while our pose tracking sustains 11.0 mm fingertip keypoint relative error during manipulation, when occlusion is heaviest. We show that these combined signals support object property estimation, slip detection, and representative recording of hand-object interaction - yielding demonstration data with the fidelity required for contact-aware manipulation policies and world models.
Embedding-level scaling expands the capacity of language models through lookup-based memory rather than by increasing model depth, width, or dense computation. Extending this idea to vision-language models (VLMs) is challenging because visual features are continuous, spatially structured, and query-dependent, unlike discrete text tokens that naturally support n-gram lookup. We introduce V-LUMEN, a visual lookup memory for VLMs. V-LUMEN discretizes visual features with a frozen codebook, builds spatially aware memory keys through spatial aggregation-based hashing, and modulates them with text-conditioned relevance signals; all addressing is non-parametric, and the retrieved embeddings are fused into the backbone through a gated residual pathway. On five reasoning-intensive benchmarks, V-LUMEN substantially improves a frozen Qwen3-VL-2B backbone and transfers zero-shot to three benchmarks whose images are disjoint from the training data. At a matched parameter budget, moving part of a LoRA or MoE adapter's capacity into V-LUMEN preserves accuracy on two backbones, indicating that external visual memory is an independent capacity axis rather than a stronger adaptation method. Because lookup cost is constant in table size and incurred only once at prefill, this capacity grows in storage rather than in per-token compute. Code is available at https://github.com/miso-choi/V-Lumen.
Large language model (LLM) agents are extending agent-mediated commerce on the Web from shopping for consumers to operating businesses for sellers. In online-to-offline (O2O) commerce, decisions are issued through Web-based platforms but realized through costly physical execution. Evaluating such agents requires going beyond individual transactions to ask whether decisions made online turn into sustained economic returns. We introduce AutoVend, a benchmark in which one LLM agent manages 20 vending machines across five site types through a Web-based operating interface for 180 simulated days. Its data-grounded simulator combines a shared warehouse, procurement lead times, paid replenishment, inventory expiration, and recurring operating costs. We evaluate five LLM agents against a rule-based inventory-control baseline and diagnose their site coordination, execution, and performance over time. Four of five evaluated LLM agents fall short of this baseline, and returns per dollar of inference cost rank models differently from returns alone. Operational analyses show that gains in individual metrics do not consistently yield stronger economic growth, as inventory and execution costs erode these gains. Trajectory analysis further reveals difficulties in grounding feasible actions, managing inventory commitments, and converting feedback into effective policy revisions.
As large language models advance rapidly, they are increasingly used to answer medical questions, and many such models are now deployed locally. Once a model is deployed, its behavior is governed mainly by the ethical policy written into its constitution. However, most existing work focuses on factual correctness rather than the trade-off between competing ethical values. Even benchmarks that do evaluate ethics treat it as a fixed property of the model rather than as a variable, so how a constitution reshapes medical-ethics behavior remains largely unexplored. We introduce MedConstBench, a medical-ethics benchmark that treats the deployment constitution as a controlled variable, is organized around Beauchamp's four principles, and is scored by human-calibrated multi-model peer review. It includes MedDilemma, our own dataset of clinical dilemmas, and supports a systematic analysis of how constitutions steer LLMs. Evaluating ten open-source LLMs under four constitutions, we find that the effect of a constitution is dimension-selective and model-dependent, and that no constitution is universally optimal. Motivated by these findings, we propose a do-no-harm adaptive selector that chooses a constitution per model, matching the per-model oracle while guaranteeing a worst-case floor and generalizing to unseen models at low cost.
Self-supervised training is monitored without labels, through the loss and statistics of the learned features. We report a case where we know the encoders lost part of what they had learned, so we can check whether these signals noticed. We trained LeJEPA encoders on synthetic images with known generating factors. In nine of ten seeds, the readability of those factors fell abruptly and stayed down, nearly always along the same direction in factor space. Meanwhile the loss kept falling, and most of the eleven label-free signals we tracked kept improving and rated the declining runs better than runs that did not decline. Standard change-point detection rarely flagged the decline near its onset, and selecting checkpoints by these signals mostly chose ones after it.
In the past few years, Vision Transformer (ViT) has achieved competitive performance in the field of computer vision. However, there still remain some problems. For instance, ViTs rely on large-scale training data since they simply contain a large number of trainable parameters, which slows down their learning efficiency and limits their applicability in resource-constrained scenarios. In this paper, we propose a novel deep learning model, namely All-Convolutional Self-Attention Network (ACSAN), which applies convolution operations to replace fully-connected matrices multiplications in the ViT framework. To implement ACSAN, we firstly design Dynamic Convolutional Self-Attention (DCSA) as the basic module, which employs multi-scale dilated convolutions to generate Query, Key, and Value. We define spatial self-attention and channel self-attention to generate attention features respectively. The two attention features are finally merged as the output of the DCSA module. Based on DCSA, we construct the basic ACSAN block. Specifically, we use DCSA to replace the regular self-attention operations, and use convolution-based Equivalent Convolutional Fully Connected (EqConvFC) modules to replace MLPs. To evaluate the performance, we train ACSAN on CIFAR-10, CIFAR-100 and ImageNet-1k datasets from scratch. Experimental results show that ACSAN achieves good classification accuracy (the best variants achieve 93.52\% on CIFAR-10, 73.38\% on CIFAR-100, and 83.3\% on ImageNet-1k). Moreover, in comparison with standard ViT and other state-of-the-art methods, ACSAN uses fewer parameters to achieve comparable or better accuracy.
We develop and analyze multilevel gradient-regularized Newton methods for unconstrained convex optimization with Lipschitz-continuous Hessians. By computing Newton corrections in lower-dimensional subspaces, the proposed methods aim to reduce computational costs while preserving the global $\mathcal{O}\left(k^{-2}\right)$ convergence rate in function values of the full-space method. To establish this rate for multilevel methods, we introduce a class of row-nested restriction operators that ensures monotonicity of the reduced-gradient norms across levels. Numerical experiments with row-nested operators demonstrate convergence behavior comparable to full-space gradient-regularized Newton, while offering substantial computational savings on large-scale instances.
The design optimization of building energy systems represents a complex, high-dimensional search problem involving discrete technology selections, continuous sizing parameters, and intricate economic trade-offs in which every candidate must be evaluated by a year-long simulation. This paper presents AutoEnergyDesign, in which a Large Language Model (LLM) agent reasons over past simulation results, proposes configuration changes and learns in context from each outcome, while a git-based mechanism reverts changes that do not improve the design. The agent couples a reasoning model with explicit memories, a constrained action interface, and an environment-grounded keep/revert feedback signal. The simulator covers eight technologies and computes the 25-year Net Present Cost (NPC). On a case study with measured building loads, the agent, instantiated with three Claude models (Haiku 5.5, Sonnet 5.5 and Opus 5.5; ten runs of 50 simulations each), is compared with random search, Tree-structured Parzen Estimator (TPE) Bayesian optimization and a (1+1) evolution strategy that shares the agent's acceptance rule (ten seeds of 300 evaluations). With all three models the agent reaches an NPC of 3.53MEUR, 1.20% and 0.23% below the best results of TPE and of the evolution strategy, and comes within 1% of it after a median of 18 to 27.5 simulations; among the baselines, only the evolution strategy does so, after a median of 192 evaluations. Every decision is recorded, making the search auditable.
LLMs on edge devices are constrained by limited memory and bandwidth. Although 1.58-bit ternary weights greatly reduce model storage, long-context inference is increasingly bottlenecked by KV-cache traffic and attention. Existing accelerators also use separate datapaths for ternary linear layers and attention, incurring hardware and execution overhead. We present SPARTA, a sparse ternary LLM accelerator for efficient long-context edge inference. SPARTA integrates: (1) a phase-aware sparse dataflow exploiting heterogeneous sparsity across compute-bound prefill and memory-bound decode; (2) a unified compute fabric mapping ternary SpMM/SpMV and bit-serial attention onto shared processing elements to exploit zero-weight and zero-bit sparsity; and (3) an adaptive two-stage attention engine combining page-level Top-k selection to reduce K-cache accesses with token-level Top-p pruning to reduce V-cache traffic. SPARTA reduces KV traffic by 4.1-4.8x over full attention and 1.3-1.5x over a state-of-the-art sparse-attention scheme, with task-dependent quality loss. Compared with NVIDIA Jetson AGX Orin, SPARTA achieves 3.9x higher generation throughput and 8.3x higher end-to-end energy efficiency on average; versus AMD Ryzen 7 9700X, the gains are 5.0x and 81.7x. It also delivers 1.6-12.0x higher throughput, 2.0-4.2x higher energy efficiency, and 1.3-51.4x higher area efficiency than FIGLUT, TeLLMe, and MCBP.