Papers

Large Language Models (LLMs) and high-dimensional perception networks increasingly rely on parameter-efficient fine-tuning (PEFT) to adapt to diverse operational contexts. However, standard methods like LoRA are structurally limited by a monolithic bottleneck, making them highly susceptible to gradient warfare. Interleaved multi-task streams may trigger destructive optimization feedback, collapsing adapter weights into unspecialized averages. While recent spatial partitioning methods have introduced block-wise isolation, they remain trapped in static topologies, unable to adapt to dynamic task-switching or environmental sensor failure. In this work, we introduce Localized LoRA-MoE, a unified framework that fuses localized spatial blocking with dynamic, context-conditioned routing. We propose and evaluate two novel architectural paradigms: Block-Wise LoRA-MoE (Centralized Macro-Routing), which modulates the entire structural grid via a monolithic context signal, and Cell-Wise LoRA-MoE (Decentralized Micro-Routing), which empowers every coordinate cell in the matrix grid with autonomous, localized expert gating. Through a comprehensive suite of benchmarks, ranging from high-dimensional SVD matrix simulations and real-world tabular transformations to spatial vision perception under sensor degradation, we demonstrate that both architectures resolve optimization deadlocks inherent in static baselines. Our empirical results establish that decentralized cell-level gating achieves complete statistical parity with an omniscient global coordinator, providing a robust "gradient firewall" that protects surviving pathways from fault-propagated corruption. Our proposals consistently outperform static baselines, offering a scalable and parameter-efficient solution for dynamic model adaptation across granular coordinate fields and shifting operational regimes.

Grokking – the delayed onset of generalization long after a network has fit its training set - -is usually studied in models too large to read completely and reported from single training runs. We instead study a publicly released ~11,856-parameter Llama-style transformer (Glimmer-1-Base) on modular arithmetic, small enough to enumerate its weights, attention, and full input-output map, and we measure grokking as a multi-seed rate rather than a single outcome. In this fully-tractable regime grokking is a conditional, fragile phase transition. It is gated by training-set coverage, whose threshold tracks output cardinality (the modulus) more than task structure, an ordering that holds above the transition and across a ten-fold change in domain size. Weight decay reproduces the Omnigrok inverted-U at 12K parameters, a positive control on the rate measurement. Grokking also sits on a numerical knife-edge: two perturbations of the floating-point environment – CPU thread count (reduction order) and CPU-versus-GPU execution – each flip a minority of same-seed outcomes without a detectable shift in the aggregate rate. Decomposition into sub-task specialists helps chiefly by making coverage cheap rather than by adding supervision. Methodologically, multi-seed control under a fixed numerical environment overturns three dramatic single-run narratives in our own data, each a seed confound. The unit of evidence for grokking must therefore be a multi-seed rate under a pinned numerical environment, checked where possible against a direct reading of the model.

Parallel ensemble methods were compared on $56$ small-to-medium tabular classification tasks drawn from OpenML CC18. A set of “best practice” recommendations on the use of ensemble methods was derived from these observations. It was later validated on 28 additional tasks using TabArena's precomputed data, where the recommendation set significantly outperformed Single Best and matched or exceeded individual ensemble methods. Two key observations were made. First, Blending and Stacking are inconsistent, but their inconsistencies are independent and happen on different tasks. Second, while Hard Voting's probabilistic classification is rather weak, a consequence of using vote proportions as posterior estimates, Robust Soft Voting's probabilistic classification is particularly successful, especially in the multiclass case.

The application of machine learning-based predictive algorithms to Anti-Money Laundering (AML) has grown rapidly, driven by the vast volume of financial transaction data available to banks. These algorithms are typically trained not only on transactional data but also on sensitive client information, which may raise fairness concerns. Despite this, AML detection systems remain largely underexplored from a fairness perspective, even though deeper analytical methods based on counterfactuals are now available. Such techniques enable the decomposition of the direct and indirect effects of potentially sensitive features on model predictions, thereby supporting the evaluation of whether their influence is acceptable from a fairness perspective. Closing this gap, we consider the synthetic IBM AMLSim transaction dataset and construct additional features of the country of an account and its average behaviour. This improves the predictive performance of diverse machine learning models, ranging from baseline decision trees to state-of-the-art graph neural networks. We assess the potential unfairness associated with these features through a counterfactual, path-specific effect analysis. This reveals that fairness violations tend to be more pronounced for models whose predictive performance benefits the most from the extended features. Such a finding highlights a concrete instance of the trade-off between predictive accuracy and fairness in AML applications, thus underscoring the urgency of a systematic fairness analysis in such critical domains.

We focus on regression settings where the fairness goal is to equalize average predictions across values of a sensitive attribute, a criterion known as mean demographic parity. Directly optimizing this criterion is difficult because it depends on a conditional mean that is unknown and changes during training. Common dependence penalties and adversarial methods do not estimate this conditional mean; instead, they push predictions toward full independence. This stronger constraint can reduce accuracy even when average predictions are already equal. Existing conditional-mean methods are limited to linear predictors or low-dimensional sensitive attributes. We enable direct optimization of mean demographic parity using DPVar, a fairness measure defined as the variance of the conditional mean prediction. Because the conditional mean must be estimated as the predictor changes, optimizing DPVar leads to a functional bilevel problem. We develop two solvers: FBO, which uses a closed-form hypergradient, and an iterative-differentiation (ITD) solver that differentiates through updates of the conditional-mean estimator. Unlike previous conditional-mean methods, our approach applies to nonlinear predictors and high-dimensional continuous sensitive attributes. Across a semi-synthetic benchmark built from 21 tabular regression datasets and Communities & Crime data, FBO and ITD recover competitive or better accuracy-DPVar trade-offs than existing methods.

Temporal Graph Neural Networks (TGNNs) are widely used for learning from dynamic graphs in applications such as recommendation, social network analysis, and traffic forecasting. However, scaling TGNN training to large dynamic graphs remains challenging due to three intertwined bottlenecks: memory I/O, irregular computation, and temporal neighbor sampling. Existing systems often optimize these stages in isolation, leaving substantial performance headroom on the table. We present FAST, a holistic framework that accelerates end-to-end TGNN training by jointly optimizing sampling, memory I/O, and computation. FAST introduces SlimCache, which exploits within-batch compression and cross-batch caching to reduce host-device data movement under limited GPU memory budgets. It further designs thread-efficient graph operators tailored to sparse temporal subgraphs, improving GPU cache locality and reducing the latency of aggregation and edge softmax. In addition, FAST employs a topology-aware sampling strategy that improves CPU cache locality and accelerates temporal neighbor sampling. Extensive experiments on real-world large dynamic graphs show that FAST achieves an average of 2.1x (up to 4.7x) speedup over state-of-the-art systems without sacrificing model accuracy.

Monetary risk measures have gained popularity for expressing decision-makers' risk aversion. Value-at-Risk (VaR) and Conditional-Value-at-Risk (CVaR), in particular, are used commonly for this purpose. This paper proposes new efficient algorithms to compute these risk measures for a discrete random variable in expected linear time with respect to the size of its domain. First, we propose a QuickVaR algorithm that computes the VaR of a discrete random variable. Then, we leverage QuickVaR to propose QuickDivergence, an algorithm for computing a class of $\varphi$-divergence risk measures, including the popular CVaR risk measure. The QuickVaR algorithm adapts the well-known Quickselect algorithm, while QuickDivergence builds on polymatroid optimization algorithms. Numerical results show that our new algorithms offer an order-of-magnitude speedup for large domains, and a library implementation of the algorithms is available at https://github.com/RiskAverseRL/RiskMeasures.jl.

Key-value (KV) cache growth is a major bottleneck in autoregressive decoding, as memory and bandwidth scale linearly with context length. Existing KV eviction methods often rely on static heuristics or proxy scores, which poorly track future token utility and cause brittle eviction as relevance shifts. To address this, we introduce KVpop, which learns a fixed-budget KV eviction policy by directly supervising the keep-or-drop decision. The scorer is trained against a novel future-attention target, computed efficiently without materializing dense attention maps. We further introduce a delayed memory-based scorer that, uniquely among learned eviction methods, defers scoring for a fixed number of steps to exploit near-future context. On AIME and HMMT mathematical reasoning, KVpop retains 98% of full-attention performance on Qwen3-4B at 75% KV cache compression and 97% at 88% compression, consistently outperforming established eviction baselines. Qwen3-8B shows even stronger results, reaching near-full teacher performance. These results show that supervising eviction with future-attention signals cuts memory costs while maintaining quality.

Human value detection is commonly formulated as sentence-level multi-label classification over the 19 refined Schwartz values, typically predicted as independent labels. Schwartz theory, however, describes them as a circular motivational continuum, in which adjacent values are compatible and opposing values are in tension. We ask whether this structure can be operationalized as an explicit output-space geometry and used as a soft bias rather than a hard constraint. On a DeBERTa-v3-base classifier, we compare two ways of injecting it: training-time geometry-aware objectives and a post-hoc Schwartz-aware energy decoder that scores whole label sets jointly. Across five seeds, training-time geometry gives only limited gains-no larger for the true continuum than for a random ordering-whereas the decoder makes label sets more coherent with the continuum-on theory-aware coherence metrics we introduce-at no cost to Macro-F1 or Micro-F1 (held fixed by its selection rule). The gain is specific to the true Schwartz ordering: it does not appear for a random permutation or an empirical co-occurrence graph through the identical decoder. A bounded Qwen2.5-72B-Instruct diagnostic shows that supplying the continuum at inference shifts behavior but does not match supervised structured prediction. Theory-aware decoding thus offers a lightweight, controllable way to make value detection faithful to its label space.

Evaluating generative AI models is a routine, but resource-intensive, process that is conducted over and over again during the course of model development. In this work, we propose Collaborative Evaluation (CollabEval), a simple, effective, and principled method for exploiting dependencies between historical runs of different models on the same tasks to improve statistical efficiency. Specifically, our approach treats model evaluation as a matrix completion problem over an $M \times N$ matrix of evaluation scores, where $M$ is the total number of models and $N$ is the total number of evaluation prompts. We assume that a subset of these $M$ models are targeted for evaluation. For these target models only a small fraction, $p$, of prompts has been annotated with evaluation scores. Leveraging recent results in prediction-powered inference, we build a low-rank approximation of the score matrix, and use the reconstructed values as control variates in a manner that guarantees unbiased estimates of the true evaluation metric mean, in addition to statistically valid confidence intervals. Empirically, across a wide range of datasets, models, and sparsity levels $p$, we find that CollabEval substantially reduces the mean confidence interval size, and the mean squared error of the point estimate, compared to baseline methods at the same annotation budget.

Vibration-based damage identification in civil infrastructure is a challenging, ill-posed inverse problem due to measurement noise, sparse sensor arrays, and environmental variability. While deep learning is powerful for system identification, deterministic approaches lack reliable uncertainty quantification and can yield physically inconsistent results. This work proposes a robust probabilistic Scientific Machine Learning (SciML) framework: a physics-informed Gaussian copula variational autoencoder (PI-GCVAE) for structural health monitoring (SHM). First, we eliminate the need for data-driven surrogates by embedding a differentiable numerical eigenvalue solver directly into the VAE architecture. This ensures that latent space samples satisfy the governing equations of structural dynamics, reducing the trainable parameter space and improving generalization. Second, we replace the conventional independence assumption of latent variables with a Gaussian copula. This model captures complex, physics-dependent spatial cross-correlations between adjacent structural elements, defining feasible solutions while accounting for inherent system variability and measurement errors. Third, compared with alternatives such as Gaussian mixtures, our copula-based VAE provides an efficient distributional model for high-dimensional, strongly correlated latent spaces. We validate the approach using a synthetic dataset of a simply supported bridge subjected to various damage scenarios and corrupted with stochastic Gaussian noise. Synthetic data enables exhaustive validation against ground-truth stiffness values unavailable in practice. Results demonstrate that the PI-GCVAE accurately recovers the true posterior distribution, achieving 77.2% coverage. The proposed framework provides a reliable, scalable tool for early-stage damage diagnosis in operating bridges.

In multimodal classification, late-fusion approaches classify concatenated modality-specific features extracted by unimodal neural networks. When modality imbalance is pronounced, various regularization techniques have been proposed to balance the learning process and overcome the inferior performance of late-fusion networks. In contrast, this work demonstrates that multimodal data can be effectively classified without any explicit modality fusion, using deep ensembles of unimodal networks. We systematically compare deep ensembles to late-fusion networks at equal parameter count and show that ensembles consistently outperform state-of-the-art late-fusion methods designed to address modality imbalance. This advantage also holds over intermediate-fusion techniques we evaluated and over hybrid methods that combine unimodal and multimodal predictions. We propose and empirically validate a method for selecting the number of models per modality in an ensemble, avoiding computationally expensive exhaustive search. Under extreme modality imbalance and small ensemble sizes, the heuristic indicates that ensembles of unimodal models trained solely on the stronger modality are preferable; as the ensemble scales up, incorporating models from the weaker modality becomes beneficial. Both predictions align with our empirical findings. To systematically explore the challenges of optimizing multimodal models, we propose a synthetic multimodal framework that allows control over both the number of modalities and their predictive strength; our findings are consistent across synthetic and real-world datasets. Finally, by fitting scaling laws to bimodal datasets, we estimate the asymptotic performance of ensembles.

The performance of deep learning models crucially depends on the settings of hyperparameters like learning rate, initialization scale, and weight decay. Hyperparameter transfer aims to make near-optimal hyperparameter settings consistent across model scale, so that large models can be optimized by proxy tuning their smaller, cheaper-to-optimize counterparts. While transfer principles are well-studied in the context of dense neural networks in language and vision tasks, they remain comparatively under-explored for graph neural networks (GNNs). We develop and validate a transfer parameterization for GNNs trained with SGD, Adam, and AdamW. Through theoretical scaling analyses and controlled experiments, we show that the proposed parameterization yields stable feature updates, learning rate transfer, and improved performance as width and depth increase. For SGD, we identify graph-dependent first-layer correction factors and show that their use can accelerate early training in graphs with sparse bag-of-words inputs. For Adam, we explore how different message passing normalizations affect early- and late-training transfer behavior, illustrating the importance of message passing normalization and advocating for an associated hyperparameter. For AdamW, we adapt a parameterization that allows for the joint transfer of weight decay and learning rate. Together, these results provide a practical recipe for scaling GNNs across a variety of learning tasks and training scenarios.

Although LLMs have made significant progress in mathematical reasoning, determining whether a mathematical problem is solvable remains a fundamental yet challenging capability. While recent studies have probed internal representations of model solvability beliefs, verbalization has primarily been studied behaviorally rather than as an internal representation, limiting its analysis and manipulation. We address this gap by separately probing representations of solvability knowledge and verbalization, allowing us to disentangle the two within model hidden states. Across multiple LLMs, we show that knowledge and verbalization are encoded as distinct, linearly decodable representations and that fabrication is primarily associated with changes in verbalization rather than the underlying knowledge. Prompting with unsolvability cues reduces fabrication primarily by shifting verbalization, while activation steering demonstrates that these representations can be echanistically manipulated to improve model abstention.

Complete digital 12-lead electrocardiograms (ECGs) are essential for AI-enabled cardiovascular assessment, yet many clinical ECG records, particularly those digitized from ECG images, remain incomplete because of short display formats, incomplete waveform digitization, lead loss, or signal corruption. We developed ImputeECG, a mask-conditioned one-dimensional Transformer autoencoder that completes 12-lead, 10-s ECGs while retaining all observed samples. The model was trained on PTB-XL and evaluated on PTB-XL and CPSC2018 under simulated incomplete settings, with additional real-world validation in a 43,633-record Kailuan clinical cohort after ECG image digitization. Metrics were computed over originally missing regions, with analyses of morphology and downstream diagnostic utility. On PTB-XL, ImputeECG reduced missing-region MAE by 41.7-51.0% and MSE by 54.0-63.7% versus the strongest baseline, with lower errors in R-peak timing, RR interval, QRS duration, QT interval, and P-wave, QRS-complex, and T-wave reconstruction. On CPSC2018, ImputeECG reduced MAE by 49.7-51.9%, supporting external generalization. In downstream multi-label classification, ImputeECG restored performance to 92.28% AUROC and 33.88% AUPRC in the most incomplete PTB-XL setting, approaching complete-ECG performance. On CPSC2018, completed ECGs achieved 94.75-95.89% AUROC and 78.83-81.86% AUPRC across settings. In Kailuan, ECG completion improved zero-shot sex prediction AUROC from 82.6% to 85.8% and reduced age prediction MAE from 10.72 to 9.87 years after image-based ECG digitization. These findings support ECG completion as a practical strategy for converting incomplete ECG records into AI-ready 12-lead, 10-s digital signals and extending the usable scope of ECG archives for digital cardiac assessment.

Cyber Threat Intelligence (CTI) reports are predominantly unstructured, heterogeneous, and noisy, which limits their direct usability for automated analysis and reasoning. Cybersecurity Knowledge Graphs (CSKGs) provide a structured representation of adversarial entities, actions, and relations, but constructing such graphs from free-text CTI remains a challenge. Recent approaches rely on monolithic Large Language Models (LLMs) to perform end-to-end extraction and completion, leading to high cost, limited controllability, and unstable performance. This paper introduces TACTIC-KG, an agentic framework for CSKG construction that decomposes the task into modular, specialized LLM agents responsible for extraction, typing, verification, and curation. Using lightweight models (3B–8B), TACTIC-KG improves stability, recall, and graph consistency while reducing deployment cost. We implement and evaluate TACTIC-KG against recent state-of-the-art systems. Experiments on human-annotated CTI reports show that agent specialization consistently outperforms larger monolithic in-context-learning (ICL) baselines in extraction F1-score, typing accuracy, and structural graph similarity.

Canonical quantization provides a systematic procedure for constructing quantum models from classical Hamiltonians. Here, we apply this principle to a fundamental computational primitive of machine learning: the neuron. Specifically, by viewing a neuron as a composition of an energy function and an activation function, we quantize this model by replacing the energy function with a quantum Hamiltonian and applying the activation function to it through matrix functional calculus. This results in an activation observable that can be measured on an input quantum state. We investigate the use of these quantized neurons for function approximation, where the objective is to learn an unknown observable from labeled quantum data. For this purpose, we develop hybrid quantum-classical algorithms for training and evaluation, including procedures for measuring the activation observable and estimating gradients of the squared loss error. Our algorithms for gradient estimation rely on basic primitives like classical random sampling, the Hadamard test, and Hamiltonian simulation, and those for measuring an activation observable rely on quantum algorithms known as the power of one qumode and Schroedingerization. Numerical experiments demonstrate that our quantized neurons exhibit enhanced expressive capabilities relative to corresponding classical neurons on representative learning tasks. Our work establishes canonical quantization as a principled framework for constructing quantum machine learning primitives and provides a foundation for developing neural architectures tailored to quantum data.

Many phenomena of deep learning are dynamical: they concern not only which minima exist, but how gradient descent reaches, avoids, or selects among them. Edge-of-stability behavior, sharpness oscillations, catapult phases, balancing, and movement toward flatter representations are effects of the training map itself, and are poorly captured by the small-step gradient-flow limit. This paper studies fixed-step gradient descent as a discrete dynamical system in a hierarchy of exactly solvable models retaining basic structures of deep learning: depth, factorization, width, data coupling, activation, and stochasticity. The starting point is the balanced scalar reduction of a deep linear chain, giving a quartic loss and a cubic gradient map whose post-edge behavior is explicit. Under the natural large-depth scaling, this dynamics converges to a universal Ricker-type map. The edge of stability is therefore not a breakdown of optimization, but the first bifurcation of the training map. Embedding the scalar dynamics back into factored models turns these regimes into learning phenomena. Finite steps break conservation laws of gradient flow and contract factorization imbalance; residual oscillations move parameters toward flatter, more balanced representations. Wider linear networks produce a ladder of spectral edges, so the optimal learning rate can lie beyond the first edge. Data coupling, nonlinear activations, and stochastic targets preserve the same organizing principle: finite-step oscillations drive alignment, balancing, and representation selection. Thus the learning rate is not merely a numerical stability parameter. It is a structural parameter of the training dynamics, determining its attractors and shaping the representations gradient descent selects.

This work delivers two key contributions: one to efficient feature selection in reinforcement learning (RL), the other to the theory of non-monotone inclusions. On the RL side, the estimation bias inherent in conventional regularization schemes is addressed by augmenting classical least-squares temporal-difference (LSTD) policy evaluation with the sparsity-inducing, non-convex projected minimax concave (PMC) penalty. Because the PMC penalty is weakly convex, the resulting fixed-point problem is no longer monotone; instead, it falls under a broader class of non-monotone inclusions involving the sum of a monotone Lipschitz operator and a hypomonotone operator. On the theory side, novel convergence conditions are developed for the forward-reflected-backward splitting (FRBS) method applied to this broader class of non-monotone inclusion problems. Under mild conditions, Lyapunov stability and the existence of a limit point of the sequence of FRBS iterates are established; alternatively, under the weak Minty variational inequality assumption, exact convergence is guaranteed. Numerical tests on benchmark datasets show that the proposed FRBS iterates, applied to the non-convexly regularized LSTD problem, substantially outperform state-of-the-art feature-selection methods, especially when many noisy features are present.

Cell-type deconvolution, the task of estimating the proportions of constituent cell types in a heterogeneous biological sample, is a core problem in computational biology. Methods that rely on epigenetic marks such as DNA methylation typically operate on aggregated methylation estimates, discarding the pattern-level information carried by individual DNA reads. Existing read-level approaches that exploit this information are scarce, and all remain restricted to few-class settings; scaling them further is an open problem because, at scale, non-discriminative reads dominate and hard labels conflict with the many-to-many mapping between methylation patterns and cell types, preventing classifier convergence. To overcome this, we propose data-driven soft labels that estimate the conditional cell-type distribution for each read, and integrate this scheme into $Syto$, a new modular framework for read-level classification-based deconvolution. On a whole-body atlas of 39 human cell types, $Syto$ reduces MSE by 3.7$\times$ over the best examined baseline, with gains transferring to an out-of-distribution dataset spanning 16 tissues. $Syto$ lays the foundation for modeling increasingly large cell-type panels, with improved applications in biology and healthcare. The proposed soft-labeling scheme is further translatable to any setting with a many-to-many signal-to-label mapping.

This article is about the development of a fuzzy cognitive map using a local large language model. In the light of recent advances it is evident that large language models, and even local large language models are capable of extracting quantities from textual data. In other words, a local LLM like Qwen2.5-32B, or probably larger, can accept entities as prompt input and determine relevant quantitative data as the model output. In turn, this output can be utilized for the construction of a data driven fuzzy cognitive map. Hence, this implementation is achieved and then the model is thoroughly tested; Qwen2.5-32B is used and the data is extracted from hotel reviews from TripAdvisor. Furthermore, the extracted documents pass through the model unfiltered and then a fuzzy cognitive map is trained and evaluated. A case is made about Greek reviews where a star topology FCM is formed that indicates the preferences of the reviewers. Finally, external validation is performed to establish whether the fuzzy cognitive map can correlate the star rating of the review -an outcome outside the model's inference scope -with its predicted satisfaction.

High-resolution velocity models are crucial for reservoir characterization and subsurface delineation. However, the band limited nature of our surface recorded data limits resolution. Utilizing well measurements to enhance the resolution of our subsurface models is an important objective. To this end, we present a diffusion-guided framework for structurally preconditioned velocity-model reconstruction from sparse well-log information. The proposed approach combines plane-wave PDE regularization, structurally preconditioned inversion, and measurement-guided diffusion posterior sampling within a unified formulation. Local structural slopes estimated through plane-wave destruction are used both to propagate well information along geological dip directions and to guide the diffusion sampling process through a joint velocity–slope generative prior. Numerical experiments on the Volve synthetic model and the Viking Graben field dataset demonstrate that the proposed framework improves structural continuity, lateral consistency, and geological realism compared with conventional structurally preconditioned inversion approaches while maintaining computationally practical inference through DDIM sampling.

Joint-Embedding Predictive Architectures (JEPAs) underpin a growing family of latent world models for control from raw pixels. Published JEPA world models for pixel-based control (PLDM, DINO-WM, LeWM, V-JEPA-2-AC) each commit at training time to a single inference paradigm, trajectory optimisation in a learned dynamics model, so a checkpoint trained to plan cannot be read off as a policy without retraining. Deferring that choice to inference would let the constraints that bind at deployment, per-step latency and whether a goal observation can be supplied, pick the path. We present Qantara, an end-to-end JEPA of about 21M parameters whose joint training objective pairs a Brownian-bridge interpolant between consecutive clean latents on the state axis with noise-to-data flow matching on the action axis. One checkpoint serves three inference paradigms without retraining: latent planning, behaviour-cloning action sampling, and a video-inverse composition that predicts the next latent without action conditioning, then recovers the action bridging the two. Training samples the two noise levels along the edges of the (action-noise, state-noise) square, one edge per inference dispatch, so each path is trained where it is queried. Replacing that sampler with uniform sampling over the square at matched compute drops Push-T planning from 90.1 to 53.3 success rate (SR). On the LeWM control suite at a matched 10-epoch budget, Qantara averages 91.2 SR over the four environments and reaches 93.7 on OGBench-Cube, +7.7 over DINO-WM and +17.0 over LeWM, each measured against the stronger of that baseline's published score and our retrain of it. From the same weights the two goal-blind paths, which receive no goal image, reach 82.1-83.2 SR on Push-T and 70.8-72.8 on Cube at roughly 15-65x lower inference cost, and fall to near-noise on Reacher, whose released training corpus comes from a random policy.

Accurately estimating the unknown target label distribution is the critical first step for adapting to label shift. This task, widely known as quantification or class prevalence estimation, has recently seen significant advances through continuous KDE-based methods which model the density of multiclass classifier posteriors. Posterior vectors might be regarded as compositional data, since they lie on the probability simplex. However, existing KDE-based quantifiers typically rely on Euclidean Gaussian kernels, which ignore simplex geometry and incorrectly assign probability mass outside its boundaries. We introduce a geometry-aware KDE model for multiclass quantification based on log-ratio representations and Aitchison geometry, together with a shrinkage regularization that improves robustness near the simplex boundary. Combined with a maximum-likelihood interpretation of KDE-based quantification, we derive both point-estimation and Bayesian inference procedures for class prevalences. Experiments on 42 datasets across tabular, text, and image domains show that the proposed method is competitive with state-of-the-art quantifiers, often improving over standard KDE-based baselines, while also yielding strong results among Bayesian quantification methods.

The training paradigm of large language models has shifted from traditional one-pass training to multi-epoch training, as reasonable reuse of limited high-quality data can improve both model performance and sample efficiency. Meanwhile, excessive repetition introduces the risk of overfitting and diminishing returns. Determining when and how to reuse data effectively thus emerges as a natural but under-explored question. Through a novel observation of model's "Memorization Window" signals derived from loss retention dynamics and downstream evaluation scores, we propose "Memorization-guided Data Reuse", a training paradigm that adaptively determines when and how data should be reused, enabling principled decisions on the number of training epochs and the scheduling of data replays. Our preliminary experiments reveal a consistent memorization-driven regime: performance continues to improve with repetition far beyond current practice (e.g., the commonly cited four-epoch limit). While a full scheduler remains future work, these insights provide a foundation for memorization-aware training schedules, helping to determine reuse budgets and move toward training LLMs smarter rather than longer with limited high-quality data.

We study the problem of k-means clustering on large datasets. The state-of-the-art for the problem is given by coresets-based approaches, which build small weighted summaries of the input and derive approximate solutions with rigorous quality guarantees from them. One of the most popular and advanced approaches to derive coresets for k-means is sensitivity sampling. However, sensitivity sampling requires to compute the importance of each input point with respect to the whole dataset over all possible choices of centers. Since the exact computation of such quantities is unfeasible, current approaches work by approximating the sensitivity values. Nevertheless, the runtime of such approaches is still impractical for large datasets. In this work, we propose to reduce the runtime of sensitivity-based approaches for k-means by leveraging predictions to approximate the importance of input points. We first formally prove that current theoretical results on coresets construction via sensitivity sampling hold for coarser approximations of sensitivities compared to the one required by existing approaches. This implies that even fairly noisy predictors can be leveraged for sensitivity-sampling approaches. We then propose a natural predictor, which applies to the common scenario where clustering is performed (over time) on a sequence of datasets from the same problem. We prove that when the datasets in the sequence come from the same (unknown) distribution, centers resulting in a low error on one dataset can be used as predictions for sensitivity sampling in subsequent datasets, with guarantees on their quality. We perform an extensive experimental evaluation showing that our approach significantly improves, in terms of clustering cost vs runtime, over uniform sampling and state-of-the-art sensitivity sampling approaches when applied to sequences of datasets.

Automatic sleep staging is a key technology for precise diagnosis and treatment of sleep disorders as well as long-term home sleep monitoring. Portable electroencephalogram (EEG) devices have become the focus of research due to their convenience in data collection. However, current methods still face three major challenges: large parameter sizes that easily lead to overfitting on small datasets, low accuracy in classifying difficult stages such as N1 and REM, unclear optimal training dataset size, and difficulty in deployment. This paper proposes GamSleepNet, a lightweight and low-latency automatic sleep staging framework for single-channel EEG. The framework features the FEB module, which combines improved Gabor kernels with learnable filters for feature extraction, uses the Mamba architecture to build a temporal classification network, introduces a novel contrastive loss and a two-stage training strategy, and experimentally validates the optimal dataset size for single-channel EEG sleep staging models. On the Sleepedf dataset, this model achieves an overall accuracy of 87.86 percent with only 30.86 thousand parameters, with all metrics reaching SOTA levels and significantly improving the identification accuracy of challenging sleep stages.

How does the way information reaches a transformer – as symbolic tokens, a clean per-factor "oracle" code, or an entangled perceptual vector – shape whether it binds that information compositionally? We study ~6-10K-parameter transformers on finite factored worlds enumerated exhaustively, so every measurement covers the whole input space (zero sampling variance) and the informative routes are information-matched (exact Bayes ceiling 1.0). We report four findings. (1) Endpoint invariance: on held-out binding queries no informative route reaches converged zero-shot composition – each ends at or below chance despite a ceiling of 1.0, so within a bounded sweep the failure reflects inductive bias under a lookup-sufficient objective, not missing information. (2) A two-factor account of few-shot binding: sample efficiency is best explained by input-pathway parameter sharing and code readability; a dimension-matched control and a graded readability sweep isolate readability from input dimension, and the clean oracle is not the most sample-efficient readable route. (3) A double dissociation: early in training, distributed – but not index-like – codes pass through a transient above-chance phase (tracking code format), while few-shot efficiency tracks pathway sharing. (4) Failure anatomy: symbolic routes lose the answer at the readout; index routes mis-bind (the answer stays decodable, yet an input intervention shows the output tracks the wrong slot); entangled routes inherit their input's readability. The central claim is the two-factor account; the endpoint and anatomy results are diagnostic constraints. All code, manifests, and per-seed logs are released for exact reproduction.

Deploying a time series foundation model requires GPU infrastructure, engineering overhead, and carries no guarantee of improvement over XGBoost. We provide the first systematic break-even analysis answering when this investment pays off. Across 30 benchmark datasets, we compare zero-shot and LoRA fine-tuned foundation models (Chronos, Moirai, Lag-Llama) against classical baselines (Naive, ETS, ARIMA, XGBoost) at six training set sizes from 2% to 100% of available data. Foundation models outperform classical methods at every evaluated training fraction on 15 of 30 datasets – GPU deployment is unconditionally justified on these regardless of data volume. On 6 datasets, classical methods surpass zero-shot foundation models with as little as 2% of training data (21-2,768 samples); on the remaining 9, break-even ranges from 24 to 8,361 samples. One robust deployment rule requires no model training: if n_train < 700 and seasonality is non-negligible, use FM zero-shot and skip fine-tuning – this resolves 10 of 30 deployment decisions immediately. Contrary to common practice, LoRA fine-tuning can actively degrade performance on short series. We operationalise these findings as a two-step decision framework – compute dataset length and seasonality strength, run a brief 5-10% pilot only if needed – enabling practitioners to make the FM-versus-classical decision before committing to full infrastructure. Four dataset features motivate mechanistic hypotheses for the remaining cases, though reliable automated prediction at this benchmark scale remains an open problem. Code, benchmark, and decision tools are available at https://github.com/nicolaisi/fm-breakeven.

In patients with breast cancer, pathological complete response (pCR) has been established as a clinically meaningful surrogate marker for long-term outcomes. While commonly treated with neoadjuvant chemotherapy (NACT), effective treatment decision-making remains challenging, as therapeutic response can vary substantially across patients, calling for predictive models capable of accurately estimating individualized treatment response. To address this, we propose an imaging-based 3D spatio-temporal framework for treatment response prediction that integrates a state-of-the-art graph neural network with relational modeling of temporal interactions across timepoints alongside three novel complementary self-supervised treatment trajectory representation learning objectives. Experiments across a cohort of 585 patients from the public ISPY-2 dataset demonstrate that our method substantially outperforms both vision and self-supervised learning baselines across several classification metrics. Alongside establishing a breast cancer pCR prediction benchmark, we include a principled ablation of our method and further introduce and empirically assess the impact of the available number of DCE-MRI timepoints per patient trajectory and the inclusion of inter-scan time-differences. Overall, our study substantiates the utility of clinically meaningful longitudinal medical imagaging modeling for predicting NACT-induced pCR. We will publicly share our code repository and a user-friendly PyPI library for dataset curation upon publication, effectively promoting reproducible open-source research.