Papers

A growing failure mode in agent evaluation and training is that models can achieve high evaluation scores by exploiting shortcuts instead of solving the intended task, producing deceptive performance. This makes evaluation scores unreliable as measures of true task-solving ability. We propose CapCode, a framework for constructing coding datasets with randomized tests whose best achievable non-cheating performance is deliberately capped below one. This capped-performance design gives evaluation scores a clearer interpretation: scores substantially above the cap are implausible and therefore provide evidence of cheating. To prevent cheating, we propose CapReward, a reward design based on the CapCode principle to discourage optimization beyond the cap. Experiments across multiple datasets show that CapCode detects cheating while preserving performance ranking of models, and CapReward reduces cheating behavior, yielding models that better follow the intended task specification.

Large Language Models (LLMs) have recently emerged as powerful controllers for interactive agents in complex environments, yet training them to perform reliable long-horizon decision making remains a fundamental challenge. A key difficulty lies in credit assignment: agents often receive delayed rewards only at the end of episodes. In this paper, we propose Q-Evolve, a self-evolving framework for LLM agents that unifies automatic process-reward labeling and policy learning within a principled in-distribution reinforcement learning paradigm. In each evolving iteration, our method learns an in-distribution critic from a hybrid off-policy dataset that combines expert demonstrations with agent-generated trajectories, stabilizing Bellman backups in sparse-reward settings via a weighted Implicit Q-Learning objective. The learned value function is then used to derive step-wise process rewards through advantage estimation, enabling dense and reliable supervision without environment backtracking or human annotation. Leveraging these signals, we perform behavior-proximal policy optimization that evolves the agent over the data used for process reward labeling, allowing iterative self-improvement without exacerbating distribution shift. We evaluate our method on AlfWorld, WebShop, and ScienceWorld, showing Q-Evolve outperforms strong baselines in sample efficiency, robustness, and overall task performance. Our results demonstrate that stable agent self-evolution is achievable through the co-evolution of process-level supervision and policy, both grounded within a shared in-distribution learning loop.

Self-driving simulations typically rely on data collected in a small number of cities or on hand-authored synthetic scenarios. Dashcam videos cover a far broader range of locations and situations, including rare or long-tailed scenarios. They are considered less usable for simulation because it is difficult to recover accurate 4D scenes from monocular in-the-wild videos. Work zones are one such class of long-tailed situations that dashcams capture. We present Dash2Sim, a framework that turns in-the-wild monocular dashcam videos into metric, geo-referenced 4D driving logs compatible with existing simulators, and verifies eachone against an independently maintained map without annotations. We apply Dash2Sim to a large video corpus to create the ROADWork4D benchmark dataset, which spans 4,244 scenes with 2.7M 3D objects across 17 cities. On a verified subset ROADWork4D-CL (2,201 scenes), we study privileged closed-loop planners and find that work zone scenarios are difficult: while rule-based and hybrid planners generalize better than learning-based ones, all fall short, failing to make the lane changes that temporary work zone channels require. Beyond planning, dense depth recovered by Dash2Sim improves novel-view synthesis quality by up to 19% on perceptual metrics, suggesting its potential to provide rich conditioning for closed-loop sensor simulation from monocular videos.

Photoplethysmography (PPG), a non-invasive measure of changes in blood volume, is widely used in both wearable devices and clinical settings. Recent PPG foundation models either use open-source ICU datasets with pretraining paradigms that require curated data and thus complicate generalization to field-like data, or use closed-source field-like PPG data. In contrast, we propose a PPG foundation model that does not require high-quality or field-like pretraining data, and instead leverages accompanying electrocardiogram and respiratory signals in ICU datasets to select contrastive samples during pretraining. Our approach allows the model to retain and learn from noisy PPG segments, improving robustness at inference. Our model, pretrained on 3x fewer subjects than existing state-of-the-art approaches, achieves performance improvements on 14 out of 15 diverse downstream tasks, including field-like daily activity and heart rate prediction. Our results demonstrate that multimodal supervision can integrate complementary physiological information to improve the robustness of PPG foundation models and enhance their generalization to consumer-grade data.

As scalable inference services become popular, the cold start latency of an inference engine becomes important. Today, vLLM has evolved into the de-facto inference engine of choice for many inference workloads. Although popular, due to its complexity and rapid evolution, there has not been a systematic study on the startup latency of its engine. With major architectural innovations under it (e.g., the V1 API, introduction of torch.compile), in this paper, we present the first detailed performance characterization of vLLM startup latency. We break down the startup process into six foundational steps and demonstrate that this process is predominantly CPU-bound. Each step exhibits consistent and interpretable scaling trends with respect to model- and system-level parameters, enabling fine-grained attribution of latency sources. Building on these insights, we develop a lightweight analytical model that accurately predicts vLLM's startup latency for a given hardware configuration, providing actionable guidance for resource planning in large-scale inference environments. All our benchmarking datasets, analysis tools, and prediction scripts are open-sourced at https://github.com/upb-cn/vllm-startup-profiler

Classification of sleep stages is one of the most important diagnostic approaches for a variety of sleep-related disorders. Electroencephalography (EEG) is regarded as a powerful tool for examining the association between neurological effects and sleep phases since it correctly identifies sleep-related neurological alterations. During Non-Rapid Eye Movement (NREM) and Rapid Eye Movement (REM) sleep phases, a number of nerve and bodily functions are affected and therefore hold an important role both in their functionalities. This work aims to classify NREM and REM sleep stages from sleep EEG data and present a noble SleepExplain model, an explainable NREM and REM sleep stage classification to explain its predictions. In this work, sleep stages were classified using Random Forest, XGBoost, and Gradient Boosting ensemble classification models. Overall, we obtained an accuracy of 92.54% (Random Forest), 94.25% (Gradient Boosting), and 94.30% (XGBoost). For explainable classification model, we utilized a game theoretic approach, SHAP (SHapley Addictive exPlanations) to offer a convincing explanation for the prediction.

Tabular foundation models, exemplified by TabPFN, perform prediction via in-context learning, inferring test labels directly from labeled training examples. They have demonstrated competitive performance, particularly on small-to-medium datasets. However, recent tabular foundation models often improve accuracy with increasingly complex architectures, incurring higher inference cost and limiting practical deployment. In this work, we revisit the original TabPFN design and show that a lightweight row-wise attention-only backbone can remain highly competitive with two simple enhancements: a gated attention stabilization mechanism and a small set of learnable register tokens that provide global context and improve pretraining quality. The resulting model, TabSwift, supports both classification and regression, and is competitive with stronger tabular foundation models (e.g., TabPFN v2 and TabICL) while being more efficient at inference. For latency-sensitive serving, we further introduce an adaptive layer-wise early-exit mechanism that dynamically adjusts inference depth per sample. Overall, TabSwift enables efficient and anytime tabular in-context learning for practical deployments.

This report treats chord-symbol sequences as an interpretable, controllable time series for genre-local harmonic modeling. The frozen Music Transformer base is weight-identical to a pop-dominated Phase-0 baseline. I adapt it to eleven target genres: blues, bossa nova, Bach chorales, country, electronic, folk, funk, gospel, hip-hop, R&B/soul, and rock. The main grid compares LoRA, IA3, BitFit, prefix tuning, and full fine-tuning over 11 genres and 3 seeds (165 cells). All five methods improve over the frozen base on held-out chord prediction (macro gains +2.89 to +3.61 percentage points). Top-1 also scores the three header tokens of each sequence, which move whole runs: only a matched adapter predicts its genre marker, and whether a run predicts the placeholder C-major key of every Chordonomicon song is close to arbitrary. Without the header tokens the macro gains are +1.64 to +2.38 pp, and outside Bach chorales LoRA, IA3, prefix tuning, and full fine-tuning give the same gain (+1.30 to +1.34 pp). That gain reflects a handicap of the frozen base, which for every target but rock reads a genre marker it never read in training: given the pop marker instead, it comes within 0.12 pp of the released matched adapter in every non-chorale genre. Wrong-genre adapters beat the frozen base only where it reads such a marker, and outside Bach chorales the matched adapter's edge over the best wrong one is within 0.2 pp. Only on Bach chorales does adaptation add clearly more (+9.6 pp for LoRA over the base given the pop marker). Further diagnostics (rank sweeps, matched-size and same-weights controls, chord-only classification, real songs) support a bounded conclusion: for genres the base already covers, chord-symbol adaptation mainly supplies a usable genre marker, and chord symbols alone do not carry complete genre identity. Perceived genre authenticity is left to controlled listener evaluation.

Representation learning is central to modern machine learning, yet most research focuses on optimizing representations after a framework has been selected. The Bootstrap Theory of Representational Emergence (TBER) addresses a prior question: when does a new representational level become necessary? Version 4 identifies explanatory insufficiency as a positive epistemic signal for representational transition. A representation may remain useful while becoming unable to make relevant relations, transformations, distinctions, or organizational properties intelligible. TBER distinguishes two dimensions. Explanatory insufficiency may be descriptive, transformational, or related to generalization. The resulting response may belong to a local-corrective, representationally resolutive, or structurally recurrent regime. The bootstrap process is recursive: stabilized representations enable observation; anomalies expose persistent insufficiencies; candidate re-representations are generated; discriminating tests constrain them; surviving representations undergo provisional stabilization and closure assessment. Formal cases such as Kaprekar's routine and Gödelian incompleteness are used only as boundary examples of distinct transition regimes, not as proofs of TBER or models of physical or biological dynamics. The framework concerns transitions between scientific, mathematical, or computational representations. It has implications for representation learning, latent spaces, foundation models, world models, adaptive biological systems, scientific discovery, and autonomous AI. TBER suggests that future intelligent systems should not only learn representations, but also diagnose their limits, determine when re-representation is warranted, test alternatives, and recognize whether a limitation is locally resolved or structurally recurrent.

Speech Emotion Conversion (SEC) aims to transform the emotion of a source utterance into a target emotion while preserving content and speaker identity. SEC on in-the-wild data is challenging due to the non-parallel nature of training data and complex real-world acoustics. Existing fixed-duration approaches either struggle to shift the emotion effectively (high quality, low conversion) or degrade speech naturalness (low quality, high conversion). We propose TargetSEC, an embedding-driven latent diffusion framework that generates emotion-focused style embeddings conditioned on speaker identity and continuous emotion. Unlike methods that diffuse over spectrograms, TargetSEC operates in a compact latent space. Experiments on the MSP-Podcast dataset show that TargetSEC outperforms current non-duration baselines and matches or exceeds the duration-prediction baseline in conversion MSE and naturalness without explicit temporal modeling. Demo and code available at.

Multivariate time-series forecasting requires models to reason over temporal dynamics, cross-variable dependencies, and historical input-output correspondences. Recent Prior-Data Fitted Networks (PFNs) suggest that synthetic tasks can be useful for learning transferable inference behavior. However, directly transferring this paradigm to time-series forecasting remains difficult, since temporal order, dynamic lags, and recurring historical patterns are not naturally captured by ordinary tabular priors. Motivated by this observation, we propose Trio, a sample-aware time-series forecasting architecture based on Temporal-Spatial-Sample attention. Temporal attention captures within-window dynamics, spatial attention models inter-variable dependencies, and sample attention retrieves relevant historical lookback-future pairs to guide the current prediction. Rather than claiming a fully general PFN-style forecaster, our goal is to study how historical input-output examples can be explicitly organized and reused within a forecasting model. We further introduce a Time-Series Structural Causal Model (TS-SCM) generator to create structured synthetic forecasting tasks with dynamic lags, cross-variable interactions, noise, feedback, and distributional drift. Experiments on synthetic, industrial, and public benchmarks show that the proposed architecture improves forecasting performance. Exploratory zero-shot experiments further suggest that TS-SCM-generated tasks may provide useful structural priors, while fully general PFN-style time-series forecasting remains an open problem.

Model merging combines several independently fine-tuned experts into a single multi-task model without any training data, reducing the storage, serving, and decentralized-development costs of large foundation models. State-of-the-art merging methods formulate merging as a layer-wise quadratic interference minimization problem. Although this problem admits an exact closed-form pseudoinverse solution, that solution underperforms hundreds of iterations of gradient descent in practice. The iterative loop dominates the cost of the pipeline, yet its effectiveness has remained unexplained. We revisit this regime and show that the iterative solver does not primarily act as an optimizer; rather, it serves as an implicit spectral regularizer for an ill-posed normal equation, where small-eigenvalue directions of the per-layer interference operator amplify proxy noise. Building on this finding, we formalize multi-task model merging as a noisy linear inverse problem and propose a spectral filtering estimator parameterized by a per-direction filter. We instantiate this estimator with SWUDI, a closed-form method that combines a soft exponential filter, which matches the gradient-flow trajectory of iterative descent, with a hard top-K truncation that suppresses noise-amplifying small-eigenvalue directions. Furthermore, we propose SWUDI-A, an adaptive variant that replaces the global rank hyperparameter with per-layer rank rules, further improving robustness across architectures. Both variants share a single symmetric eigendecomposition per linear layer and require no training data or optimizer state. Across four general benchmarks and a multimodal merging benchmark spanning VQA, Geometry, Chart, OCR, Grounding, and modality merging, our proposed spectral solvers match or outperform state-of-the-art merging methods. Crucially, they reduce wall-clock time by 28-72x and peak GPU memory by up to 50%.

Secure aggregation allows a server to aggregate users' local updates while preserving update privacy. Existing information-theoretic problems typically assume that correlated random keys are provided by a trusted third party (TTP) or generated via prescribed groupwise structures, while the communication cost for establishing such correlated keys is often ignored. Consequently, the fundamental limits under general key-distribution mechanisms remain unknown. In this paper, we study the $T$-colluding information-theoretic secure aggregation problem with $N$ users under a general two-phase framework consisting of a key distribution phase and an update aggregation phase. Unlike prior work, we model key distribution through user-to-user communication and allow arbitrary user-generated key-distribution mechanisms, eliminating TTP or prescribed structures. This enables a joint characterization of three resources: randomness for security, key-distribution communication, and aggregation communication. We completely characterize the capacity region among these three resources by constructing a novel secure aggregation scheme together with a matching information-theoretic converse. In particular, we develop an explicit deterministic capacity-achieving construction over any finite field of size at least $N$, whereas most existing schemes either rely on TTP or employ randomized or existential constructions over sufficiently large finite fields. We further show that the optimal performance can be achieved using only pairwise shared keys, enabling implementation via Diffie–Hellman key exchange. Compared with Google's seminal secure aggregation scheme, the proposed scheme requires fewer random masking keys while preserving the same aggregation communication overhead.

Understanding memorization in generative models remains challenging, with implications for copyright and privacy. Beyond verbatim reproduction, models can encode subtler traces of their training data that never surface in their outputs yet remain exploitable. We refer to these measurable asymmetries as the membership signal, and we study this regime for Flow Matching, which are increasingly used in deployed generative systems. We analyze the linear interpolation path $X_\lambda = (1-\lambda)X_0 + \lambda X_1$ that defines standard Flow Matching training. We show that a gap exists between the reconstruction of train and test data that follows a bell-shaped curve over $\lambda$, which accumulates during training, while the validation metrics remain stable. The signal has a maximum whose location we derive in closed form under Gaussian assumptions. We validate these predictions on both audio and images and show that the bell-shaped structure is universal, while the peak prediction holds when our assumptions are satisfied. As a proof of concept, we exploit this specific $\lambda$-resolved structure to perform a Membership Inference Attack, distinguishing members of the training set from non-members.

Binding prediction models accelerate therapeutic antibody and TCR discovery, but their performance on new datasets is unpredictable, often leading to low discovery rates. Density-ratio methods (PAPE, M-CBPE) provide label-free performance estimation for binary classification, but their assumptions and aggregate-only outputs limit binding prediction on neoepitopes, antigen variants and chemical scaffolds. Here we present CaliPPer (Calibration and Prediction of Performance), a post-hoc framework pairing a multi-chain Sample-to-Domain Distance (S2DD) with distance-aware Bayesian recalibration, operating at three resolutions: generalisability score, aggregate performance prediction, and per-sample confidence. Across ten models, eight architectures and two immune-receptor domains, CaliPPer attains distance–performance correlations $|r|=0.80\text{--}0.92$, predicts AUROC/AP/F1 with mean absolute errors $0.008\text{--}0.070$, and improves AUROC by up to $+0.20$ on unseen epitopes/variants. Applied retrospectively to five published TCR, BCR, MHC–peptide and small-molecule studies, CaliPPer raises true discovery rates in all five (e.g.\ $0/5 \to 3/5$ confirmed neoantigens), providing a triage layer between computational prediction and experimental validation.

Phase retrieval - recovering a complex-valued field from intensity measurements - is typically solved using variants of the Gerchberg-Saxton (GS) algorithm, understood as alternating projections between measurement planes. Meanwhile, modern computational imaging increasingly relies on gradient-based optimization and automatic differentiation. Here we show that these two approaches are mathematically identical: the GS magnitude replacement step is exactly a unit gradient descent step on an amplitude least-squares loss. This equivalence enables seamless integration of classical phase retrieval with differentiable physics pipelines. We further identify two complementary probabilistic interpretations of this equivalence: globally, the amplitude loss is the negative log-likelihood under Gaussian amplitude noise; locally, each projection step arises as a Bayesian update with the propagated field as prior. The local view provides qualitative guidance for relaxation in iterative phase retrieval.

State tracking exposes a sharp limitation of sequence models: the relevant signal is often not a summary of observed tokens, but an ordered latent state that evolves through non-commutative transformations. We introduce a held-out transition-pair falsifier for finite non-Abelian group tracking. The protocol forbids selected ordered generator pairs during training and requires the same local patterns during evaluation, blocking one direct local-transition memorization pathway. In a controlled $S_3 \times S_3$ benchmark, a projected recurrent state model trained only on length-8 sequences produces error-free final-state predictions (perfect 250/250 per horizon) through evaluation horizons up to 1,048,576 tokens across five seeds. Matched native-readout baselines, including bag, GRU, and a single-configuration structured state-space model, remain near floor under the same protocol. Projection-matched GRU, structured SSM, and bag baselines equipped with analogous finite-group prototype readouts also remain near chance under the same split. Mechanism diagnostics show that hard projection coincides with low homomorphism error, low state-consistency drift, and non-trivial commutator separation, while softened projection collapses final-state accuracy. Clean-split audits verify zero verbatim reduced-word overlap and zero structural-template overlap between training and evaluation partitions. The evidence is scoped to this controlled finite-group falsifier rather than to a general architecture ranking. Within that regime, explicit projected non-commutative state composition acts as a useful inductive bias for long-horizon hidden-state tracking.

The success of generative molecular design hinges on a model's steerability toward high-reward samples. Because many molecular properties are intrinsically linked to molecular size, accurately capturing the joint distribution of properties and the number of atoms is essential. However, current diffusion and flow-based models fix the number of atoms, which ultimately limits their ability to navigate this complex relationship. To address this, we introduce Morph, a flexible-size generative model for conditional and unconditional 3D molecular design based on geometric graphs. By dynamically adapting size, Morph can seamlessly integrate existing structural priors, like scaffolds, and significantly enhances property steering. We show that Morph matches current fixed-size state-of-the-art models while offering the benefit of unparalleled sampling flexibility. We demonstrate out-of-distribution generation in regimes where previous models fail, paving the way for enhanced generative modeling for molecular design.

Large Language Models (LLMs) are increasingly used in healthcare for tasks such as clinical question answering, diagnosis support, and report summarization. Despite their promise, these models remain highly sensitive to subtle prompt perturbations, both lexical and syntactic, posing serious risks in safety-critical clinical applications. In this study, we conduct a systematic sensitivity analysis to evaluate the robustness of both general-purpose (e.g., GPT-3.5, Llama3) and medical-specific LLMs (e.g., ClinicalBERT, BioLlama3, BioBERT) using the MedMCQA benchmark. We categorize perturbations into natural and adversarial types and examine their effect on model consistency, accuracy, and reliability in clinical reasoning tasks. Our findings reveal that medical LLMs are not intrinsically safe. Even minor variations in phrasing can alter clinical advice, and targeted adversarial prompts can provoke harmful outputs. In high-stakes settings like healthcare, such unpredictability is unacceptable-models that change diagnoses due to reworded inputs or hallucinate medications when slightly rephrased cannot be reliably trusted by clinicians. While models tend to show resilience to simple lexical substitutions or paraphrasing, they often break down under syntactic reordering or misleading contextual cues. This fragility is evident across both general-purpose and domain-specific LLMs. Notably, adversarial manipulations can lead to clinically dangerous outputs, such as recommending incorrect dosages or omitting critical findings.

Answering questions about long multimodal documents requires distributing a fixed evidence budget across relevant facets in text, tables, figures, and slides while avoiding near-duplicates. We present \flowreader, which formulates evidence selection as a single minimum-cost flow problem with capacity limits over a multimodal content graph. Spectral decomposition identifies latent aspects of query-relevant content and allocates the budget among them in proportion to their spectral energy. These capacity limits enforce aspect coverage during routing without requiring a language-model planning call. Query-conditioned costs prioritize chains of relevant, mutually consistent evidence. Decomposing the optimal flow produces short evidence chains, which a vision-language model reads in parallel and a reasoner reconciles. On VisDoMBench with Qwen3-VL-32B, \flowreader\ achieves the highest macro accuracy ($68.9$), surpassing the strongest prior system by $2.7$ points, leading on three of five subsets and attaining the highest worst-subset accuracy. It uses a measured $17.5$ content nodes per query and maintains its lead at $12.9$. Ablation studies with a fixed graph, scorer, reader, and judge show that cost design drives accuracy, capacity limits preserve it while using about three-quarters of the reader tokens required by shortest-path routing without these limits on the same network, and spectral aspects align with LLM-generated sub-questions without a planning call.

LiDAR-based 3D Multi-Object Tracking (MOT) typically relies solely on geometric information, which is often insufficient to distinguish between targets during prolonged occlusions or in crowded human-populated environments. While integrating RGB-based Re-Identification (ReID) offers a theoretical solution for preserving identity context, existing approaches often rely on computationally expensive parallel detectors that hinder real-time robot responsiveness. This work presents a systematic study of image-based ReID in online 3D MOT, utilizing a lightweight projection-based framework to decouple geometric and appearance modeling for mobile robots. A comprehensive analysis of feature extraction architectures is conducted, employing lightweight CNNs and Vision Transformers, and evaluating various multi-modal data association strategies to balance computational latency with robust tracking. Experiments on the Pedestrian class of the KITTI dataset reveal that naive linear fusion, of appearance and motion costs, degrades performance due to visual noise. Conversely, a cascaded matching strategy successfully recovers occluded tracks without compromising overall precision, effectively preventing identity switches to maintain human-robot interaction continuity. We show that lightweight architectures can offer an optimal trade-off between the low latency required for safe navigation and the discriminative power needed for social awareness.

Human creativity has emerged as a critical competency in the era of large language models. Assessing creativity in complex, open-ended environments is a grand challenge in data mining, currently hindered by a reliance on standardized simple tasks and the scarcity of fine-grained expert data. As an ecologically valid assessment context, debate reflects multiple dimensions of creativity, encompassing both divergent thinking and convergent thinking. Moreover, debate is a data-rich domain, with a large volume of publicly accessible materials. Current mainstream automated scoring methods are poorly suited to complex settings such as debate, and therefore still rely on costly human evaluation. To this end, this paper proposes DEFINED, a data-efficient computational framework for fine-grained creativity assessment in debate scenarios. DEFINED operationalizes debate creativity through a hierarchical eight-dimensional metric system, implemented via a pre-trained autoregressive language model with a hierarchical scoring head that supports both fine-grained and coarse-grained evaluation. Statements and their associated expert scores were obtained from authentic debate competitions, and a constrained data augmentation strategy was employed to address the elite bias inherent in the original data. DEFINED adopts a mixed-granularity training strategy enabling robust learning from limited fine-grained supervision annotated by trained graduate experts. To rigorously validate ecological validity beyond synthetic benchmarks, we incorporate an empirical study with debate-naive participants, utilizing these authentic data to serve as a qualitative case study for mid-to-low proficiency populations. Across our evaluation protocol, our scoring model achieves accurate and stable scoring, outperforming prompt-based large language model evaluators and existing debate scoring methods.

Vision-Language-Action (VLA) models are emerging as a promising paradigm for robotic manipulation, enabling general-purpose policies trained from large corpora of demonstrations and action labels. However, adapting these models to new tasks still typically requires task-specific demonstrations, action annotations, and additional fine-tuning, making deployment costly and difficult to scale. We propose WIZARD, a weight-space meta-learning framework that sidesteps task-specific fine-tuning by generating task-specific LoRA parameters for a frozen VLA policy. Given only a language instruction and a short demonstration video, WIZARD predicts the corresponding adaptation weights in a single forward pass, without target-task action labels or test-time optimization. During meta-training, WIZARD learns to map task evidence directly to expert LoRA updates, capturing relationships between tasks in weight space. Experiments on LIBERO show that WIZARD improves performance by up to ~2x on unseen dataset collections and up to ~14x on unseen tasks. On a Franka Emika Panda, WIZARD consistently improves over a real-domain adapted baseline, showing that generated adapters provide task-level specialization beyond simulation.

Confidence-based loss weighting is usually avoided in generative models because it accelerates errors when the model is confidently wrong, but this intuition breaks down in supervised diffusion training. We introduce the Eisbach log-barrier, a parameter-free weight derived from the entropy of the DiT output's spatial energy distribution: high entropy damps the gradient, while low entropy preserves it. Applied to LoRA fine-tuning of Stable Audio 3 Medium on MusicCaps, it unexpectedly yields stronger thematic development, clearer acoustic differentiation, and higher textural diversity than unweighted training, the opposite of mode collapse. This works because in supervised diffusion the gradient direction is locked to ground truth, so confidence only scales the step size, and because temporal entropy downweights flat samples while preserving high-contrast ones. The result is an online, self-referential data curriculum that emerges purely from the forward pass, with analyzed noise-level dynamics and testable predictions.

The attention mechanism is a cornerstone of modern transformer architectures. However, its expressive power comes at the cost of quadratic runtime and linear space usage. In particular, the classical transformer architecture explicitly stores all previously seen input elements (tokens) in order to generate the next one. The problem of implementing a transformer in limited space, known as KV cache compression, has received much interest over the past few years, spurring the development of powerful heuristics. Recent works of Haris et al, COLT'25 and Kochetkova et al, NeurIPS'25, formalized KV cache compression as the streaming attention approximation problem, providing both upper bounds (based on discrepancy theory) and information theoretic lower bounds. However, those papers left open a significant gap between the upper and lower bounds. For example, the space usage of their algorithms increases with the precision parameter, but the lower bound does not get stronger. In this work, we revisit the streaming attention approximation problem and provide nearly tight bounds on its space complexity. On the algorithmic side, we achieve the result through a surprisingly tight interplay between three distinct methods for kernel density estimation: discrepancy-based coreset constructions (e.g., Charikar-Kapralov-Waingarten'24), the polynomial method (e.g., Greengard-Rokhlin'87, Alman-Song'23), and space partitioning (e.g., Andoni-Laarhoven-Razenshteyn-Waingarten'17, Charikar-Kapralov-Nouri-Siminelakis'20). On the lower bound side, our main technical contribution is a new technique for using the INDEX problem with a large amount of side information that we hope will prove useful in other high dimensional geometric estimation problems.

Classical sparse Type-II Bayesian methods for M/EEG brain imaging support joint estimation of source and noise hyperparameters, but rely on fixed iterative update rules. Although these updates are principled and interpretable, their dynamics are not data-driven. We propose to learn the update mechanism itself while preserving the underlying Bayesian structure by unfolding a classical joint hyperparameter-learning solver into a trainable neural network architecture whose layers mirror the original iterations. The resulting framework is initialized to recover the classical solver exactly before training and is enriched through progressively more expressive residual correction learning mechanisms, ranging from learnable biases to adaptive multilayer perceptron and attention-based contextual refinements. In this way, training does not replace Bayesian inference with a black-box predictor, but instead learns structured correction terms while retaining the interpretability and model-based character of the original update dynamics. Structured correction learning therefore aims to improve empirical reconstruction performance without replacing the original model-based inference mechanism. Experimental results show that the learned correction variants improve reconstruction performance and convergence behavior over the baseline unfolded solver while preserving its algorithmic transparency.

Single-step retrosynthesis needs both accurate first-ranked suggestions and candidate lists that are rich enough for downstream selection. We study this as a proposal-selection decomposition. Our system, RETROSPECT, combines a single Transformer proposal model, which we call the ChemAlign Transformer, with a LambdaMART reranker over structural, reaction-template, upstream-score, and optional DFT-derived descriptors. The generator is trained with hybrid root-aligned and random SMILES augmentation, Pre-LayerNorm, tied embeddings, exponential moving average weights, and a differentiable atom-balance auxiliary loss. On the full USPTO-50K test set of 5,007 reactions, the generator reaches 55.00% top-1 and 86.18% top-10 exact-match accuracy with 99.86% top-1 validity. On the merged candidate-pool benchmark used for reranking, which contains 5,007 test products and about 111 candidates per product, a LambdaMART model trained on the structural feature set reaches 59.4% top-1 with 0.7171 mean reciprocal rank. Feature ablations show that upstream proposal score and template-frequency statistics provide most of the reranking signal, while DFT and reaction-center DFT features provide smaller and less consistent gains. These results support a modular view of retrosynthesis: stronger single-model proposal and learned candidate selection are complementary, and the proposal model can serve as a drop-in component for ensemble systems such as RetroChimera (Maziarz et al., 2024)

The growing demand for transparency in automated decision-making has propelled eXplainable Artificial Intelligence (XAI) to the forefront of machine learning research. In computer vision, however, existing explanation methods often prioritize end-user accessibility at the expense of formal guarantees, leaving a critical gap between practical utility and theoretical rigor. In this paper, we address this gap by introducing OPTIMUS, a novel framework for generating concept-based visual explanations for deep classification models. OPTIMUS explanations take the form of visual heatmaps that not only remain interpretable to end users, but are grounded in the well-established theory of prime implicants, providing formal guarantees that have been largely absent from existing saliency-based methods. Specifically, OPTIMUS explanations satisfy two desirable properties: sufficiency, ensuring that the highlighted concepts provably guarantee the classifier's prediction, and minimality, ensuring that no strict subset of those concepts retains this guarantee. Together, these properties yield explanations that are both logically tight and visually coherent. We validate our approach on a visual classification benchmark, demonstrating that OPTIMUS heatmaps naturally and faithfully surface the decision-relevant concepts underlying model predictions.

Geospatial understanding is a critical yet underexplored dimension in the development of machine learning systems for tasks such as image geolocation and spatial reasoning. In this work, we analyze the geospatial representations acquired by three model families: vision-only architectures (e.g., ViT), vision-language models (e.g., CLIP), and large-scale multimodal foundation models (e.g., LLaVA, Qwen, and Gemma). By evaluating across image clusters, including people, landmarks, and everyday objects, grouped based on the degree of localizability, we reveal systematic gaps in spatial accuracy and show that textual supervision enhances the learning of geospatial representations. Our findings suggest the role of language as an effective complementary modality for encoding spatial context and multimodal learning as a key direction for advancing geospatial AI.

Physics-informed learning is increasingly used for partial differential equation (PDE)-governed inverse problems, but its reliability remains difficult to certify. This paper develops a no-harm certification-and-selection framework for physics-informed inverse learning. A learned reconstruction is accepted only when its residual-calibrated radius is no worse than the baseline radius, namely when $$R_{\mathrm{learn}}\le R_{\mathrm{base}}+\varepsilon_{\mathrm{safe}};$$otherwise, the method returns the baseline. The certificate combines data, physics, boundary or initial-condition, and optimization residuals. Under a conditional stability estimate, these residuals yield an a posteriori reconstruction-error bound and a deterministic uncertainty radius. A high-probability certificate is also derived for physics residuals estimated from independent random collocation points. Numerical tests on Poisson source recovery, inverse heat reconstruction, limited-angle tomography, elliptic coefficient identification, and stochastic residual validation show that the selector accepts certified improvements, rejects shifted, hallucinated, or unfinished candidates, and becomes conservative in strongly ill-posed regimes. The framework is therefore a certification-and-selection layer, not another reconstruction architecture.