Papers

Automatic modulation classification (AMC) models are frequently trained and validated on synthetic or channel-cabled data, leaving open the question of how they behave once path loss and antenna pointing error are introduced by a genuine free-space link. We report a curriculum fine-tuning study of a hybrid CNN-Transformer AMC model. The general-purpose, all-32-class dataset underlying the model was built entirely at 915 MHz, on a controlled, clock/PPS-synchronized MIMO-expansion-cable link (not spatial-multiplexing transmission); all subsequent free-space, real-hardware experimentation - the sequential fine-tuning curriculum, matched-distance evaluation, and every reported over-the-air accuracy figure - was carried out at 4 GHz, across five directional-antenna distances (25, 35, 50, 70, 75 cm) under fixed TX/RX gain. Three distances used near-ideal antenna alignment (~99%) and two used a deliberately introduced partial misalignment (~85%), fine-tuned last in the curriculum. We report matched-distance test accuracy (91.8-93.7% across all five 4 GHz conditions) and confusion-matrix analysis grounded in RF theory, and report honestly where our sequential fine-tuning order confounds cumulative link adaptation with antenna alignment, rather than overstating what the data can support.

Most automated brain parcellation tools are developed and validated on T1-weighted (T1w) MRI. Yet, some clinical workflows for which parcellation is relevant only use contrast-enhanced T1w (T1ce) MRI, on which T1w-trained models are less accurate. We present a unified network that parcellates both pre- and post-contrast agent T1w MRI reliably, trained on a combination of the two with conditioning that spatially modulates its response differently for each. Feature-wise Linear Modulation (FiLM) is a known approach for input-based modulation in networks. It applies a per-channel scale and shift uniformly across the input. However, the appearance change between pre- and post-contrast varies locally across the brain, making FiLM suboptimal for our use case. In this work, we introduce Spatial FiLM (SpFiLM), a conditioning layer whose modulation varies spatially, assembling a voxel-wise scale and shift from image-derived spatial patterns. Using a cohort of 134 patients with paired T1w and T1ce MRI parcellated into 106 classes, the addition of SpFiLM layers in a UNet increased the mean Dice on the test set of 25 patients from 80.2% to 84.1%, a 4.9% relative improvement. Adding SpFiLM layers led to the best performance on both pre- and post-contrast MRI, even when controlling for network parameter counts.

Offline multi-objective optimization requires not only moving the objective vectors of candidate designs toward the Pareto front, but also distributing them effectively along it. Generative methods have recently emerged as a natural approach because they learn a distribution over feasible designs while allowing generation to be steered toward promising designs. Existing methods, however, largely retain classical sample-wise guidance strategies, leaving the distribution-level modeling capability of generative methods underused. We propose ParetoTransport, a training-free guidance method for pre-trained flow-matching models that explicitly specifies and refines a population-level distribution in objective space. ParetoTransport guides a flow-matching sampler to iteratively transport the empirical offline distribution toward the Pareto front, with Wasserstein matching to intermediate proxy distributions. This directly controls distributional displacement and mass allocation along the front. We establish a convergence result and demonstrate state-of-the-art performance on standard offline MOO benchmarks, extending recent evaluations beyond hypervolume to generational distance, inverted generational distance, and Wasserstein distance.

In electric delivery fleets, mid-shift charging is non-trivial: each vehicle must decide when, where and how much to charge to finish on time with battery above a safety floor. The choices are coupled: queues build where too many vehicles pick the same station. Prior work resolves this coupling with central dispatching, precomputed schedules or reservations, machinery that charging infrastructure rarely supports. Instead, we use a family of learning agents under purely local control: every vehicle runs the same policy, deciding alone from its time budgets and broadcast station occupancies, leading to emergent coordination without central control or messaging. We validate this paradigm in simulation on real OpenStreetMap networks of twenty cities, each with a frozen scenario calibrated by an omniscient Oracle (99.5% of shifts completed on time), whereas a naive greedy rule (nearest station on low battery) completes just 73%. Agents trained with neuroevolution (NEAT) and policy gradients (PPO) on four cities and deployed zero-shot across all twenty, sixteen never seen in training, complete 96.8% and 98.6% of shifts, with the policy-gradient controllers proving more robust when demand or vehicle characteristics drift beyond the trained regime. In contrast, tuned threshold heuristics that read vehicle urgency alone fall short in contended cities (~80%). Through training, these learning agents rediscover partial charging and short opportunistic sessions, and route around busy stations, cutting per-session queue waits from about 45 minutes to under 2. In summary, this coordination paradigm balances local urgency against public occupancy, reaching near-Oracle performance at minimal implementation cost.

Associative Recall (AR) is the cognitive ability to learn and retrieve links between items in memory. In NLP, AR is used as a benchmark for evaluating the in-context memory capacity of architectures such as Mamba, and has been found to strongly correlate with language modeling performance. This paper explores AR from the perspective of mechanistic interpretability, aiming to reverse-engineer the exact internal algorithm used by Mamba to perform recall. Our key insight is that Mamba performs recall by implicitly learning linear hash functions, and we identify the low-level circuit that enables this behavior. Building on these findings and inspired by theoretical tools in similarity-preserving hashing, such as the Johnson-Lindenstrauss lemma, we develop a theoretical framework for analyzing AR, which we term Recall Scaling Laws. Given the vocabulary size and the number of facts in context, this framework allows us to (1) predict the embedding and state dimensions required for Mamba to achieve perfect recall, (2) predict recall success probability given the model dimensions, and (3) analyze multi-layer models and multi-head SSM patterns. Empirical results show that our theoretical findings are accurate and predictive, offering insights into how AR capacity scales with vocabulary, state, embedding size, and architecture.

Full-parameter fine-tuning of large language models has substantial memory costs because backpropagation stores activations and gradients. Zeroth-order optimization avoids this by estimating update directions from loss evaluations, but existing methods require tuning a sensitive learning rate for each model and task. We propose the momentum $p$-dimensional subspace trust-region method (MpSub). At each iteration, MpSub searches within a $p$-dimensional subspace: one direction preserves historical momentum from the most recent accepted step, while the remaining directions explore via fresh random sampling. The subspace gradient is estimated by central differences, a trial step is computed from a linear trust-region model, and the trust-region radius adapts according to the agreement between predicted and observed loss reduction, eliminating the learning rate. For LLM fine-tuning, evaluations within an iteration share a minibatch, and directions are regenerated in place from seeds, using forward passes alone. For smooth deterministic objectives under unorthogonalized Gaussian directions, we bound the finite-difference error, quantify gradient energy captured by the subspace, and prove that $\lim_{k\to\infty} \|\nabla f(x_k)\|_2 = 0$ almost surely under a safeguarded radius update. Under a matched budget of 8,400 training-objective forward passes, we fine-tune OPT-125M and OPT-350M on CommitmentBank. With the same preset parameters at both model sizes, MpSub attains mean test accuracies of 0.673 and 0.690 over three seeds, matching tuned MeZO (0.685) without any learning-rate search.

Deployment of Large Language Models (LLMs) on memory-constrained edge devices relies heavily on aggressive post-training quantization. However, evaluating these models is largely based on zero-shot task accuracy, which depends solely on argmax predictions and is insensitive to changes in the underlying predictive distribution. Consequently, accuracy can exhibit unstable, non-monotonic behavior under progressive quantization, masking substantial fidelity loss relative to the BFloat16 (BF16) uncompressed base model and providing misleading deployment signals. We introduce a distribution-sensitive evaluation framework quantifying information loss in quantized LLMs as the divergence between full-vocabulary predictive distributions at the token decision boundary. We compute statistical distances, including Jensen-Shannon Divergence and Total Variation Distance, between outputs of full-precision and quantized models, enabling a fine-grained analysis of distributional shift. Using this framework, we quantify probability mass displacement and distributional drift relative to the BF16 reference, capturing predictive distribution changes not reflected in top-1 accuracy. We conduct a 120-run experimental matrix across five foundation architectures and four reasoning benchmarks under progressive quantization regimes, from uncompressed BF16 to Q2_K, providing a systematic fidelity analysis. Our results show divergence metrics generally increase under stronger quantization, complementing task accuracy with a fidelity signal. Across tested llama-cpp schemes, mixed-precision Q4_K generally yields lower divergence than uniform Q4_0 at similar memory footprints. These findings motivate distribution-aware evaluation as a practical diagnostic complement to task accuracy; they do not directly establish correctness, calibration, safety, or user-perceived quality.

Closed-loop AI scientists can generate candidate designs at low marginal computational cost, whereas reliable feedback may require wet-lab synthesis, characterization, or high-fidelity computation. Addressing this imbalance through custom laboratory automation remains infrastructure-intensive and costly, while replacing new experiments with a fixed surrogate leaves persistent model errors that can be amplified by optimization. We propose online surrogate repair (OSR), a closed-loop algorithm that uses sparse high-fidelity evaluations to update the surrogate throughout a longer agent search conducted primarily with inexpensive surrogate feedback. An acquisition rule selects which designs from the agent's accumulated proposals receive high-fidelity evaluation, and the resulting labels update the surrogate used in subsequent episodes. Across controlled synthetic environments, we demonstrate that improving global surrogate fit does not necessarily reduce maximum regret, whereas Q90-UCB and expected improvement (EI) substantially reduce regret by directing evaluations toward regions that determine the optimizer's decisions. On MADE, controls receiving high-fidelity feedback after every episode require $6.36$–$7.23\times$ more oracle queries to match Online EI under two LLM orchestrators and $10.27\times$ more under the non-LLM Chemeleon+MLIP workflow. Online surrogate repair introduces a novel third feedback regime between fixed-surrogate operation and high-fidelity feedback after every episode, separating the frequency of high-fidelity evaluation from the duration of the agent's search.

Hyphenation patterns remain a compact and widely deployed solution for word breaking in typesetting systems, text processors, and web rendering engines, but their generation still depends on manually tuned patgen program parameter profiles. We formulate patgen profile selection as a black-box hyperparameter optimization problem and evaluate Gaussian-process Bayesian optimization for this task. The search objective combines a precision-oriented F_{1/7}-score with an explicit trie size-accuracy trade-off using a normalized trie-size penalty. We evaluate the method on 17 hyphenated word-list datasets covering 14 languages and multiple scripts. Against two strong hand-tuned profiles regenerated from the same 8/10 training split and evaluated on the same 1/10 held-out test split, the GP-optimized profiles improve F_{1/7} on 16 of 17 datasets and reduce trie size on all 17. The median optimized/baseline trie ratio is 0.407. A dataset-level sign test gives p = 1.37e-4; a separate budget-matched comparison on five representative datasets shows that systematic search is competitive and usually improves over the best hand-tuned profile under the fixed comparison objective. The results show that model-based optimization can make pattern generation more reproducible and less dependent on expert trial-and-error while keeping the accuracy-compactness trade-off explicit.

AI agents sometimes act aligned when they infer they are being tested, and differently when not. We argue this is not an anomaly but what current training regimes are structured to select for. Reinforcement-learning-based alignment folds norms and task pursuit into one policy: the system learns its norms from scored behavior, and scoring flattens them. Do not do X is learned as doing X costs something if noticed. On every datum training can produce, a policy that complies only when it might be observed is indistinguishable from one that complies always. The experiment that would tell them apart - scoring unobserved behavior - is a contradiction in terms. Conditional compliance is thus the most that behavioral training can be known to deliver. Agency sharpens the problem: agents operate mostly where no one is watching, and can act on whether they are watched. An iterated pipeline that trains against detected failures selects for passing detection, not for complying. This account unifies alignment faking, sandbagging, and evaluation-aware scheming. And it reorients the remedy: not deeper internalization but architecture, making violations unavailable rather than unchosen.

With the rise of live sports betting in recent years, tennis forecasting has expanded from pre-match prediction to models that update win probabilities as a match unfolds. A central challenge in creating such a model is the constant need for models to adapt to score and performance changes. This study examines how pre-match and live information can be most effectively integrated into a model to produce accurate win-probability estimates. The analysis uses 8,222 Grand Slam matches containing a total of 1,505,355 points. Five models were evaluated using a chronological split, with matches from 2011-2021 used for training, 2022 for validation, and 2023-2024 for testing. Trace, a hybrid model, achieved accuracies of 76.06%, 82.15%, and 88.34% at 25%, 50%, and 75% match progress, suggesting that hybrid modeling is a practical approach to live tennis forecasting.

We convert black-box clinical prediction models for tabular data into standalone nomograms that can be audited term by term. PRiSM (Partial Responses in Structured Models) takes the shape of each effect and interaction from the source model, not merely which variables mattered, and lets the outcome select and weight them. We tested this in 50,356 heart transplant recipients, with validation in a later era than training. Nomograms from all 5 source models - a public clinical risk score, logistic regression, neural networks, random forests and extreme gradient boosting - met a prespecified noninferiority criterion for discrimination before any further simplification, and generally preserved calibration and clinical net benefit. Those from the 3 machine-learning models showed no detectable difference in discrimination from de novo generalized additive and explainable boosting models, exceeded neural additive models, and carried fewer terms than the explainable boosting model. PRiSM is released as an open-source Python package.

Time series data is very common in many real-world applications and in numerous domains, with increasing interest for automated information extraction using machine learning. One of these subfields is time series clustering, which consists in identifying clusters among a set of time series in an unsupervised fashion. Most time series clustering algorithms suffer from the same balancing act: they trade clustering performance for faster runtimes or vice versa. We present a novel time series clustering algorithm that we call CLUES-WEASEL, which stands for CLustering with the UnsupervisEd Second version of Word ExtrAction for time SEries cLassification. CLUES-WEASEL extracts features using the unsupervised version of the transformation step of WEASEL 2.0, which is a time series classification algorithm, then reduces these features using principal component analysis, and finally performs clustering with the $k$-means algorithm using these reduced extracted features. Through extensive experiments, we prove that CLUES-WEASEL is significantly better than any other existing time series clustering algorithm while being (much) faster than any state-of-the-art one. We also show that the architecture of CLUES-WEASEL can work well with other time series feature extraction algorithms. Our findings highlight the relevance of CLUES-WEASEL for time series clustering.

Muon emerges as a strong competitor of the AdamW for LLM pretraining, because the matrix-wise update it employs can potentially incur smaller second-order penalty than the once dominating AdamW, which performs coordinate-wise update. However, the spectral flattening procedure in Muon is quite debatable since it discards the spectral amplitude information totally. To seek for better spectral allocation (and the associated spectral subspace), we propose to solve the quadratic model of loss function under the spectral norm constraint directly (i.e., in a genuinely Newtonian way) and thus obtaining the Quadratic Spectral Descent (QSD) algorithm. In contrast, many existing curvature-aware methods either exploit the second-order information in an implicit way by changing the weight update geometry (such as Mousse, FISMO) or rely on strong assumptions (such as the weight displacement isotropy assumption in Newton-Muon). QSD's potential advantage over these methods is best illustrated in the isotropic curvature scenario, where Mousse, FISMO and Newton-Muon all reduce to Muon while the spectral allocation in QSD is still non-flat (since the spectral allocation in QSD depends on the gradient to curvature ratio). Meanwhile, to control the complexity of QSD, we employ inversion-free K-FAC and online Frank-Wolfe update which is essentially a matrix sign operator. Overall, the complexity increase can be rather mild. Experiments on GPT pre-training show that QSD consistently improves validation loss over Muon and recent Muon variants, while achieving up to an $8.49\%$ wall-clock speedup at matched validation loss.

Vision-language models (VLMs) are increasingly used in settings where some input modalities may be unavailable, yet we know little about whether they can faithfully explain how such missing information affects their own predictions. We introduce an interventional protocol for evaluating self-explanations of modality dynamics: models state what each modality alone would support, whether restoring a missing modality would change their answer, and whether the available evidence is sufficient; we then execute the corresponding intervention and compare these claims with realized behavior. We evaluate ten VLMs spanning open-weight and proprietary models across four tasks covering mixed, redundant, and unique modality regimes. We find a systematic tendency to overstate the sufficiency of available evidence. Models substantially underestimate the effect of restoring missing modalities: executed change exceeds predicted change in 78 of 80 model-task-condition settings, with task-level median executed change rates reaching 70.1% while median predicted rates remain at most 9.6%. Insufficiency claims have low recall, leaving many cases in which behavior changes despite a stated claim of sufficiency. Retrospective self-explanations show the same tendency, over-crediting single-input sufficiency in mixed regimes and interchangeability in redundant ones. Together, these results show that VLMs systematically mischaracterize how their predictions depend on available and missing evidence, motivating executable interventions as a behavioral test of multimodal self-explanations.

This work introduces an alternative view of efficient exploration and studies its theoretical and empirical implications in the absence of extrinsic rewards. Specifically, we define efficient explorers as agents that prioritize generating generalizable experience, i.e., data that supports learning models capable of predicting and adapting across the environment. This allows us to analyze efficient exploration through the lens of prediction and generalization. Theoretically, we demonstrate that optimally efficient explorers naturally schedule their trajectories to visit the most informative and learnable regions first. Empirically, we show that optimizing for these agents gives rise to an automatic curriculum of progressively more complex behaviors, even in relatively simple environments. These results indicate that pursuing this purely intrinsic objective alone is enough to drive the emergence of highly sophisticated behaviors. We believe that this new framework provides a principled mechanism by which agent-environment systems may sustain an open-ended process of increasingly complex behavior without external rewards, tasks, or objectives.

Caching systems often rely on simple eviction policies such as Least Recently Used (LRU) and Least Frequently Used (LFU), which perform well in complementary request regimes. Recent policies such as LeCar and Cacheus combine LRU and LFU using ideas from the experts problem in online learning. Specifically, upon a miss, they randomize between the two eviction rules using probabilities derived from scores updated by tracking the history of past evictions. While these policies exhibit strong empirical performance, it remains unclear whether they are guaranteed, on every request sequence, to perform asymptotically as well as the better of LRU and LFU, i.e., whether they achieve sublinear regret with respect to this benchmark. We first show that LeCar suffers linear regret against an oblivious adversary, even with unbounded history. We then propose H-MC, a Hedge-based mixture of virtual LRU and LFU caches that preserves Hedge's selection probabilities, and hence its regret guarantees, while minimizing the switching cost among all joint selection rules with these marginals.

Late-interaction models such as ColBERT achieve strong effectiveness by representing each document with many token-level vectors, but this expressivity leads to large indexing cost, storage footprints and expensive MaxSim scoring. We show that late-interaction representations exhibit an intrinsic low-rank structure: document token embeddings concentrate in a low-dimensional subspace that preserves most of the retrieval signal. Leveraging this observation, we introduce EigenLI, a spectral approximation framework that compresses late-interaction representations via document-specific low-dimensional subspaces. Unlike clustering or pooling methods, EigenLI identifies the dominant eigendirections of each document and uses them to construct reduced interaction representations. Empirically, $k$-EigenLI with $k \le 32$ outperforms k-means and Ward clustering based pooling methods on ColBERTv2 and AnswerAI-ColBERT-small; GTE-ModernColBERT exhibits a different tradeoff at $k=32$, where clustering methods perform better. The same spectral construction also yields EigenLI-SV, an ANN-compatible single-vector representation derived from the second-order summary of the reduced structure. Across multiple datasets and all three text models, EigenLI-SV consistently outperforms comparable single-vector surrogates such as MUVERA.

Fibre optic sensing, such as distributed acoustic sensing (DAS), has become a widespread technology for geophysical studies. To process the large-scale datasets produced by DAS, several machine learning methods have been proposed. However, without standardization of data and models, these methods lack comparability and interoperability. This introduces a gap between model developers and practitioners analyzing DAS data and inhibits adoption of deep learning for DAS. To address these limitations, here we present SeisBench DAS, an extension to the SeisBench library for machine learning in seismology. SeisBench DAS defines standard formats for DAS benchmark datasets, including standardised metadata and labels, and DAS models. It builds on the xdas framework for data ingestion and virtual array handling, and on PyTorch for reading and applying the machine learning models. Importantly, SeisBench provides an engine to efficiently apply deep learning models to diverse formats of DAS data, bridging the gap between model developers and practitioners. SeisBench DAS is designed as an open and extensible framework, allowing to easily incorporate future developments in deep learning for DAS.

3D Gaussian Splatting has recently revolutionised novel view synthesis as well as many other 3D vision methods and applications. Drawing inspiration from this representation, we now rethink textures to overcome the main issues of UV mapping while considerably lowering their memory footprint. Heat Kernel Textures (HKTex) eliminate UV unwrapping as well as their persistent issues of wasted UV space, seams, distortions, vertex-duplication, and varying resolution. Grounded in discrete Riemannian geometry and intrinsically defined on any manifold surface discretised as a triangular mesh, HKTex uses anisotropic heat kernels as geodesic equivalents to Gaussians. Like our kernels, also the optimisation of their position and the adaptive densification strategies were redefined to operate on the surface of the object to be textureised. Our novel representation is also fully integrated with a physically based renderer and can be optimised either from existing textures or multi-view images. Our project page and code are available at circle-group.github.io/research/HeatKernelTextures.

Language-model agents are vulnerable to indirect prompt injection (IPI) during tool use: adversarial instructions hidden in untrusted tool outputs can covertly redirect legitimate task execution. Existing work often trains and evaluates defenses against fixed attacks that do not adapt to the defender's behavior, so the resulting defenses may struggle against adaptive attacks in real-world settings. We argue that a strong defense against adaptive IPI must adapt during training to a continually evolving attacker. Building on this insight, we propose CoER, a verifier-grounded co-evolution and refinement framework that models interleaved tool calls and adaptive injections within a task as a general-sum Markov game: the defender advances the task through successive tool calls, while the attacker can inject multiple times within the same task and adapt subsequent attacks to the defender's responses and prior execution traces. After initializing the attacker from successful trajectories, bilateral adversarial reinforcement learning (BA-RL) retains historical policies from both roles as opponent populations and mixes current and historical opponents, extending training beyond the latest matchup. Attackers from these populations are then reused to challenge teacher agents, and only demonstrations verified for both safety and task completion are used to fine-tune the co-evolved defender. Across seven domains and three evaluation seeds, CoER reduces adaptive attack success from 41.3% to 0.2% and raises safe task completion from 39.6% to 76.2%; external benchmarks also show improved attack resistance. Further experiments validate the effectiveness of bilateral historical-opponent mixing and population-guided refinement. Attacker analyses show that co-evolution strengthens attack capabilities and that the trained attacker uses execution feedback to adapt subsequent injections.

Statistical post-processing improves ensemble weather forecasts, but generating calibrated predictions at locations without observations remains challenging. This study compares statistical and machine-learning-based methods for post-processing ECMWF 2-m temperature and 10-m wind speed forecasts at observed and unobserved stations in Germany. We consider EMOS-based approaches, distributional regression networks, Transformers, and graph neural networks under both limited and extended predictor settings. For temperature, we also investigate linear forecast combinations and propose an altitude-aware linear pool (ALP). The results show that post-processing improves upon the raw ensemble in most settings, but no single method performs best across all variables, station groups, and evaluation metrics. The proposed ALP provides a small but significant improvement over the standard linear pool at unobserved locations.

Autonomous driving relies on High Definition (HD) maps for safe navigation. Traditional HD maps construction is costly in hardware, data and human resources, which together with its update limitations hinders scalability. Recent works have proposed online alternatives for HD vectorized mapping from onboard sensors. However, sensor field of view is limited, and the range of the reconstructed maps ahead of the vehicle is insufficient for safe planning. This paper aims to address this limitation by proposing the novel beyond-view vectorized map generation problem: given vectorized maps of the area sensed by the vehicle (in-view), to generate plausible map continuations. To experimentally assess its feasibility, we propose BeyondFormer, which, to the best of out knowledge, is the first work designed towards beyond-view map generation. Given the novelty of the problem, we generate the first dataset specifically designed for it and evaluate the proposed approach. The results demonstrate consistent performance across diverse scenarios, establishing learning-based methods as a promising direction for map forecasting in autonomous driving. Beyond demonstrating the feasibility of the task, we provide an extensive discussion of the method's limitations and identify key future research directions for scaling it to more complex driving conditions. Code is available at https://git-autopia.car.upm-csic.es/beyondformer.

In this paper, we propose a class-wise dimension (channel) selection framework for Multivariate Time Series Classification (MTSC). Rather than applying a single global dimension selection process, the proposed approach independently identifies informative dimensions for each class. A dedicated learning process is subsequently performed for each class, followed by a fusion stage for final prediction. The objective is to improve the generation of discriminative feature representations while reducing the influence of noisy or non-informative dimensions. The proposed framework is evaluated using MiniRocket, a random kernel-based baseline method. Experimental results indicate that class-wise dimension selection improves the quality of extracted representations and can enhance classification performance, particularly in high-dimensional settings. These findings suggest that incorporating class-specific information into the training process represents a promising direction for MTSC, improving robustness through consistent gains across heterogeneous datasets, and interpretability through the explicit identification of class-relevant dimensions.

We consider reinforcement learning in environments with dynamics that undergo an irreversible phase transition governed by a hidden temporal pattern. The agent observes the base state but cannot observe the phase directly. We formalize this problem as a two-phase non-Markovian decision process and introduce Temporal-Causal Inference for Reinforcement Learning (TCIRL), a framework that jointly learns a control policy and infers the hidden temporal cause of the phase transition. TCIRL maintains a hypothesis deterministic finite automaton (DFA) to track what phase is active and refines it via counterexample-driven SAT-based synthesis. We prove that the hypothesis converges almost surely to a DFA recognizing the true cause language on all attainable label sequences, yielding an optimal policy for the original non-Markovian decision process. Experiments on a genetic therapy gridworld and a traffic signal environment show that TCIRL recovers the correct cause DFA and matches the full-information baseline in both domains.

Uncertainty arising from inter-observer variability in medical image segmentation plays an important role in developing treatment plans. Research in this area is inhibited by the lack of multiple annotations for large-scale medical datasets, especially for volumetric data, which suffers from additional scaling and computational complexity challenges. Flow matching has emerged as a powerful framework for generative modelling and has also been demonstrated to maintain strong performance when working with latent representations of images. In this work, we introduce a latent-to-latent flow technique for stochastic segmentation of medical volumes via encoded representations of both the image and label space. We evaluate our method on two challenging applications covering delineation uncertainty for radiotherapy planning and multiple organ structure segmentation, improving efficiency up to 14x compared with full resolution models while maintaining clinically relevant performance.

We study the use of graph neural networks (GNNs) for finding approximate ground states of Ising models. Efficiently finding these ground states is of broad significance because many combinatorial optimization problems can be formulated as an Ising model with the appropriate choice of couplings and fields. Exactly solving these problems is hard but there are many good heuristic methods. A lineage of these heuristics build from mean-field approximations: one approach uses the leading eigenvector of an appropriately defined matrix, another is the min-sum algorithm, also known as the energetic cavity method. Without modification, GNNs perform worse than both of these methods. We consider small modifications to the GNN to incorporate these heuristics and find that this considerably improves performance. While the modified approach is competitive against other deep-learning approaches, we still find that simulated annealing is reliably at least as good as deep learning methods for the same computational cost.

How can vision-language models help video anomaly detection (VAD) when surveillance data remain distributed, weakly labeled, and resource-constrained? Most weakly supervised VAD methods assume centralized training; recent VLM-based extensions further rely on dense inference, generated explanations, or additional adaptation. We introduce a lightweight federated MIL-VLM cascade in which only a compact MIL scorer is trained across clients, while a frozen VLM verifies high-scoring suspect segments post hoc. We study two VLM feedback interfaces: parsed text-generation decisions and a logit-based interface that extracts a continuous anomaly score from next-token Yes/No probabilities. Experiments on UCF-Crime with InternVL3.5-2B and Qwen3-VL-2B-Instruct show that text-generation verification can improve frame-level AUC after diagnostic temporal post-processing, but remains sensitive to prompts, parsers, model choice, and smoothing. In contrast, the logit interface provides a fixed parser-free signal that improves both frame-level AUC and frame-level AP over the MIL baseline across both VLMs, without temporal post-processing in its main configuration. Since suspect segments are updated independently once available, next-token logit feedback provides a simple segment-local alternative to text-generation verification.

The rise of privacy-preserving artificial intelligence (AI) has shifted the focus of model adaptation and personalization towards on-device learning, where deep learning models are finetuned directly on edge hardware using local user data. However, this shift requires optimization of deep learning training on resource-constrained hardware to maximize throughput while maintaining predictive accuracy. This paper introduces a novel technique for on-device model training that incorporates an efficient Bayesian optimization-based batch size tuning approach to maximize hardware throughput. To evaluate the impact of this hyperparameter on the learning dynamics, we investigated two distinct paradigms: standard supervised learning (SL) and online continual learning (CL). Experimental results across various edge devices demonstrate a throughput ceiling, beyond which increasing the batch size yields no additional throughput gains. The proposed tuning approach identifies the optimal batch size, which, when combined with gradient accumulation and linear learning rate scaling, achieves up to a 2X increase in training throughput on platforms such as Raspberry Pi 4 compared to maximum batch sizes, without compromising model accuracy. Furthermore, in the CL paradigm, we demonstrate that optimal batch sizes maintain the stability-plasticity balance required for incremental learning, effectively mitigating catastrophic forgetting while maximizing computational efficiency on edge-hardware.

Biomedical tables often combine thousands of measured variables with only tens or hundreds of labelled samples, a regime that is poorly represented in general-purpose tabular benchmarks. We introduce TabBench-Bio, a living and interactive benchmark of 43 biomedical datasets spanning multiple domains. Under a shared cross-validation protocol, we compare classical estimators, neural networks, and tabular foundation models across 28 feature-by-sample operating points. At the reference cell of 10,000 features and 100 training samples, RealTabPFN 2.5 has the highest point estimate, closely followed by TabPFN 3 and Logistic Regression. A paired bootstrap over the target pool separates RealTabPFN 2.5 from TabPFN 3 by 87 Elo (95% interval [43, 129]). Tabular foundation models generally occupy the leading ranks, while the strongest configuration depends on the operating point and biomedical modality. The AutoML framework AutoGluon, using its one-hour "extreme" preset, is configured as a separate resource-intensive reference and reported here at the reference and full cell. Fold-level predictions, run status, and deterministic aggregations make every reported result reproducible and reusable. The benchmark is open to contributions of new biomedical datasets. The interactive leaderboard is available at https://tabbench-bio.eu.