Papers

Ship fuel consumption (SFC) prediction supports vessel operation optimisation, emissions estimation, and decision support systems (DSS) for sustainable maritime transportation. Numerous data-driven fuel models have been developed over the past two decades, but a critical and often overlooked limitation lies in their validation practices: most studies evaluate performance using random train–test splits, which, applied to high-frequency records, admit temporal leakage and yield optimistic results that do not reflect deployment conditions. This paper examines that gap using time-aware evaluation, specifically Time Series Cross-Validation (TSCV) and Blocked TSCV (BTSCV). Using the Canadian Coast Guard Ship (CCGS) Sir Wilfrid Laurier as a case study, six regression models and a physics baseline are tuned under three time-aware schemes and three feature configurations, then evaluated on a common chronological hold-out set drawn from approximately 3.88 million steady-state 1\,Hz records.

Video world models approximate the stochastic distribution of physical outcomes through generative sampling, but existing benchmarks score individual generations or compare distributions coarsely over a whole dataset, leaving the fine-grained aleatoric uncertainty of specific phenomena untested. We introduce CaliBench, which scores outcomes in a physically interpretable discrete space - a bin index, a die face, a suit, a colour - rather than a learned feature space such as in FID, so the distance from a known reference distribution is measured directly. We curate outcome spaces whose reference is known in closed form (binomial Galton boards, Bernoulli forks, uniform dice/cards/lottery, a skewed European-roulette colour), enabling an exact calibration test. We decompose performance into two orthogonal axes that a single accuracy metric conflates: scorability, the fraction of generations yielding a scoreable outcome, and calibration, the total variation distance from the reference on that sample. A chi-squared test assesses significance; as calibration is its null hypothesis it can evidence only miscalibration, and at N=32 per cell detects only large deviations. We apply it to nine scenes and six image-to-video models (WAN-2.7, SeeDance-2.0, HappyHorse-1.0, Veo 3.1, Runway Gen-4.5, Cosmos3-Super), 32 generations each. Models consistently concentrate probability mass on a few outcomes rather than reproducing the reference. Most scene-model combinations are significantly miscalibrated, in the extreme collapsing to one outcome, as Veo 3.1 does on dice. On roulette, generations often leave the ball ambiguously placed, giving several models low scorability. Performance varies by scene: no model dominates all nine. We release the protocol and a metric (mean normalised total variation, mnTV) for comparing new models against our results.

Generative Engine Optimization (GEO) modifies web content to increase its likelihood of being selected and cited by generative search engines. This can give strategically optimized pages visibility disproportionate to their authority or relevance and even make weak or false information appear well supported. Unlike conventional search, generative search synthesizes information into direct answers rather than presenting competing sources, which can further amplify these risks, as assessing source provenance and authority requires additional user interaction. Despite these concerns, systematic methods for detecting GEO-optimized webpages remain underexplored. We introduce GEOFlagBench, a benchmark of 3,200 web content instances spanning 400 queries, four domains, and eight GEO optimizer families, and use it to systematically evaluate existing GEO detection methods. Although the strongest baseline achieves an aggregate F1 of 0.880, method-level and authorship-conditioned evaluations reveal substantial weaknesses and potential reliance on authorship-related shortcuts. We therefore propose Intervention-Paired Training (IPT), which supervises detector responses to GEO interventions and non-GEO AI polishing; on ModernBERT, IPT improves F1 from 0.862 to 0.944 and worst-group accuracy from 0.725 to 0.883. We develop a GEO-gated Agent system for auditing the Source Tier and verifiability of Citation URLs in detected GEO pages. Finally, we deploy the complete pipeline on released Google Search and Gemini-grounded retrieval results for 1,000 real-user queries. Across 10,095 available pages, we estimate an overall GEO prevalence of 8.90%, reaching 16.36% among pages modified in 2026. Our results establish a foundation for systematically detecting, auditing, and measuring GEO in real-world search ecosystems.

Identifying and representing object instances such as cells or nuclei is a common task in microscopy image analysis. Established machine learning workflows typically use supervised detection or segmentation followed by feature extraction or classification, which requires manual annotations and treats instance segmentation and cell representation as separate stages. We describe a new unsupervised method for cell instance segmentation and phenotypic classification from unlabeled microscopy images. Our method is based on reconstructing each image using a coarse-to-fine routing pyramid that associates pixels with spatially sparse latent sources. The resulting pixel-to-latent associations yield instance masks, while the source latents encode cell morphology. We demonstrate competitive performance in instance segmentation across diverse cell morphologies and imaging modalities, as well as generative modeling of cellular phenotypes under perturbations. Source code and checkpoints are available at https://github.com/weigertlab/routing-pyramids.

Agent harnesses have substantially improved performance on long-horizon tasks by coordinating agent interactions with the environment. However, reinforcement learning through complex harnesses remains largely unexplored, as scaling such training to long-horizon agent tasks introduces fundamental challenges. In this work, we present a unified black-box RL framework for stable and scalable optimization of general agents through complex harnesses. Concretely, we first build a sandbox-based execution infrastructure that isolates task environments and harnesses within temporary sandboxes for large-scale concurrent rollouts. We then decouple policy optimization from opaque harness execution and place a serving proxy at the model boundary to capture model calls. To reconstruct multi-turn trajectories and improve training efficiency, we organize the captured calls into prefix trees and further adapt both critic-based PPO and critic-free GRPO to optimize over the recovered tree structure. Meanwhile, we maintain training-inference consistency throughout the optimization process. Finally, we introduce mix-harness training, allowing a single model to be jointly optimized by heterogeneous harnesses. With Qwen3-30A3B, black-box RL improves Pass@1 on ClawGym-Bench by 9.98 and 14.81 points through OpenClaw and Claude Code, respectively, while remaining stable over 200-400 optimization steps. Moreover, the framework yields consistent gains on more challenging tasks such as JobBench and OfficeQA. Overall, our framework enables effective, stable, and scalable optimization of general agents through black-box harnesses, supporting unified training across heterogeneous execution systems.

Prototype-based neural networks are hailed as interpretable-by-design architectures. Recently, Abductive Latent Explanations (ALE) were introduced to provide formal, mathematically guaranteed explanations that leverage the intrinsic structure of these networks to ensure both predictive safety and human readability. ALEs rely on computing tight bounds on latent space distances to produce formal explanations. However, existing ALE formulations are rigidly confined to Euclidean latent spaces. This leaves a critical gap: modern state-of-the-art architectures increasingly rely on non-Euclidean representations - such as spherical metrics, Gaussian densities, and dimensional projections - rendering current formal explanation methods incompatible. In this work, we generalize the ALE framework to support non-Euclidean prototype architectures. For each geometric variant, we systematically derive how to either map the architecture to existing bounds or construct novel, architecture-specific bounding algorithms. We validate our theoretical constructions by computing subset-minimal formal explanations on fully trained image classifiers. By unifying these diverse models under a single formal framework, we enable the first rigorous, cross-architecture comparison of their interpretability.

Reliable optimization is central to neural network (NN) training, yet Adam, the default optimizer for modern LLMs, rests on a fragile foundation. This thesis develops a principled grounding for Adam and motivates new designs. First, we revisit Adam's divergence–convergence debate and show the existence of a problem-dependent phase transition: with properly chosen, batch-size-dependent hyperparameters, Adam converges, whereas under small-$\beta_2$ regimes it can diverge. Second, we investigate why Adam substantially outperforms SGD on Transformers through Hessian structure. We find that the Hessian evolves toward a near-block-diagonal form along training, accompanied by strong block heterogeneity. We prove that this structure makes Adam's diagonal preconditioner effective. We further show that this special Hessian structure originates from consecutive multiplications of large matrix variables, and we provide a rigorous analysis based on random matrix theory. Finally, these insights motivate Adam-mini, a new optimizer that reduces Adam's memory footprint by 50% while preserving its performance. Our results also have broader implications beyond Adam: they reveal new local structures in matrix-based nonconvex problems, and also help understand and improve recent NN optimizers, such as Muon.

Many areas of AI research, such as language model interpretability and chain of thought faithfulness, seek to explain model behaviors. But what constitutes a "good" explanation? In this work, we evaluate explanations through the lens of counterfactual simulatability-whether the explanation is useful for predicting model behaviors on related counterfactual inputs. To this end, we introduce CHIVE (Counterfactual Hypothesis Investigation Via Edits), a novel agentic pipeline that identifies unexpected model behaviors in the wild and investigates them with counterfactual prompt edits. This yields thousands of high-quality explanations for naturally-occurring model behaviors along with supporting counterfactual evidence. We apply CHIVE in two ways. First, we evaluate whether common LLM interpretability techniques improve an agent's ability to predict counterfactual model behaviors. Surprisingly, we find no uplift from any of the interpretability techniques studied. Second, we use CHIVE to generate training data. We find that training models to predict outcomes of CHIVE-generated counterfactual experiments generalizes to various out-of-distribution settings. Overall, CHIVE automatically discovers explanations of naturally-occurring LLM behaviors, enabling us to evaluate and improve methods for explaining LLM behaviors.

Reinforcement learning algorithms for Large Language Models (LLMs) are largely distinguished by their variance reduction strategy. Group-relative methods like GRPO reduce gradient variance by sampling multiple rollouts per prompt, but provide only sequence-level credit. Training is also blocked by straggler rollouts, reducing throughput and increasing off-policyness. Learned value functions theoretically address both problems, providing token-level advantages without requiring large groups. However, additional infrastructure engineering challenges combined with the practical success of critic-free methods have made it difficult to justify their inclusion in RL pipelines. We propose two complementary strategies to improve the performance of value function RL: 1) Privileged Value Functions (PVF) which provide an elegant mechanism to inject additional task-relevant token-level signal without biasing the policy objective; 2) TETHER, a baseline that adaptively interpolates between group-relative and value baselines depending on the value function accuracy. Across several reasoning tasks, both strategies consistently improve over the standard value function baseline, and are competitive with or outperform mean-baseline GRPO.

As autonomous vehicles (AVs) approach Level 4 and Level 5 operational capability [SAE International, 2018], their on- board decision systems must handle not only safety-critical locomotion but also their subsequent moral weight. This paper details the Ethical Decision Head (EDH), a deep re- inforcement learning (RL) framework that encodes ethical reasoning as a differentiable reward signal, enabling a pol- icy gradient agent to learn morally-aligned driving behavior in scenarios whose state representation is aligned with the CARLA simulation environment [Dosovitskiy et al., 2017]. Two normative frameworks are instantiated and evaluated: a Utilitarian framework minimizing total casualties and a Kan- tian framework enforcing course maintenance as a categori- cal imperative. The EDH is trained via Proximal Policy Op- timization (PPO) [Schulman et al., 2017] against a Bradley- Terry reward model [Bradley and Terry, 1952] learned from pairwise human preference annotations over 200 collision- imminent scenarios. Results reveal an asymmetry in the learnability of normative ethical frameworks under human su- pervision. The Kantian condition, which reduces to a con- stant prediction task under the codebook, serves as a pipeline control: it confirms training stability and rules out infrastruc- ture failure as an explanation for the utilitarian result. The Utilitarian agent learned something more unsettling: human raters rewarded self-sacrifice over casualty minimization, and the model learned that preference faithfully. This divergence between what humans prescribe in theory and what they re- ward in practice suggests that RLHF does not learn ethics as philosophers define it, but as humans live it.

A radiologist reading a model's output faces two problems. The model returns a number and no reason, and any system that turns that number into readable prose can quietly add claims the model never made. MIRROR is a research prototype built to separate those failures. It chains a multi-label classifier, a Grad-CAM localizer that turns each positive finding into a named anatomical region, and a report writer that receives the labels, probabilities, and regions but never the image. Because the language layer cannot see pixels, it cannot assert a finding the classifier did not make. We are precise about what that buys: a MIRROR report's findings are auditable against the probability vector, while the sentences framing them are ordinary generated text, and we show one stating a cardiothoracic ratio the system never measured. One registry holds the taxonomy, anatomy, and phrasing for chest X-ray, brain MRI, and head CT, so adding a modality is a data change; all three are routed and tested, one is trained. On ChestMNIST that classifier reaches macro AUROC 0.729 and ranks better than chance on all 14 labels, at 1.6 to 6.8 times the precision a random ranker would get. Yet at the default 0.5 threshold it emits no positive prediction at all for 11 of them, and its excellent-looking Brier score of 0.045 sits beside the 0.047 earned by a predictor that ignores the image. The discrimination is real; the decisions are not. Under the class imbalance normal in radiology, aggregate metrics flatter models that do nothing, and should be reported against that floor.

Various machine unlearning techniques have been developed in response to privacy legislation requirements, enabling individuals to exercise their legal right to have their data $D_f$ removed from a machine learning model. This process is typically accomplished via the use of an unlearning function denoted as $U$. Existing methods focus on designing an intricate $U$ to unlearn $D_f \subset D$ from a previous model $A(D)$, so that the unlearned model performs as closely as possible to the retrained model $A(D \setminus D_f)$. However, these methods often suffer from high computational costs when dealing with massive training data, as the complex structures of $U$ become a bottleneck even for models with fewer parameters. Inspired by Learning to Optimize, we introduce the first learning-based model-agnostic approach, Learning-to-UnLearn (L2UL). Our core insight is to shift from manually designing $U$ to learning the unlearning behaviors from a distribution perspective, thereby acquiring a simple and efficient $U$ via learning. Our experimental results demonstrate that the accuracy achieved by L2UL is comparable to that of retraining while exhibiting impressive efficiency, particularly in data-intensive scenarios. Furthermore, we validate the performance and scalability of our method on larger models ResNet.

Motivated by modern marketplaces, where the platform or the seller routinely gathers detailed user profiles, we study a novel learning theoretic model that simultaneously involves information and mechanism design. Specifically, we consider the economic setting recently introduced by Bergemann et al. (2022), where in addition to the menu of quality-price pairs, the seller offers information on the value of the match between product quality and buyer's taste via a signaling scheme. We relax the assumption that the seller knows the buyers' belief about the distribution of tastes and study the sample requirements of designing a revenue maximizing scheme. We consider both the batch setting where we have access to data from a set of i.i.d. buyers and an online demand query model where we observe the buyers' behaviors to seller's schemes. Despite the apparent non-convexity of the problem, we also give the first FPTAS to compute a scheme that maximizes the revenue within an arbitrarily small additive loss, which was left open by Bergemann et al. (2022). Overall, this brings a new learning perspective in asymmetric economic settings where buyers and sellers know different types of information.

Lossy compression is conventionally driven by a task-agnostic distortion (e.g., MSE or MS-SSIM), yet in many emerging applications the receiver cares not about uniform fidelity but about a downstream task whose relevant content varies across the signal and evolves over time. We formulate task-aware compression as a weighted rate-distortion problem, in which a single codec is driven by a separable, per-component weighted distortion whose weights encode task importance and may depend on the source. We introduce task consistency, i.e., that minimizing the weighted distortion also minimizes the true task loss, and characterize when it holds: for linear tasks, the task loss admits a weighted-MSE form with signal-independent weights under suitable cross-term conditions, while for nonlinear tasks, an integrated-gradients analysis motivates separable task-aware weights. We show how task symmetry and irrelevance further constrain the admissible weights. Guided by this theory, we realize the weight-conditioned code in a single learned Vision Transformer (ViT) codec whose token-level attention natively consumes a per-component importance vector, so one fixed backbone is re-targeted at runtime, from universal (task-agnostic) to task-specialized operation, purely by swapping the injected weights, without retraining, while producing a single human-viewable reconstruction steered to the active task. On downstream face-analysis tasks, a single model reaches 91.4% accuracy at 0.034 bpp on a localized task, within 1.9% of a task-specific codec (93.3%) and well above a universal codec (76.9%). Such task-adaptive compression suits bandwidth-constrained perception systems, e.g., in Physical AI, where the active task drifts and per-task retraining is infeasible.

Instance-wise feature selection is a valuable tool for interpreting labeled data and the predictions of black-box models. In contrast to global feature selection techniques, instance-wise methods dynamically identify important features for each instance. A growing number of methods learn a selector, which identifies important features, and a predictor, which uses these to make predictions. However, these pioneering methods face challenges including information leakage and lack of differentiability, which can slow training. In this paper, we present Hide&Seek, an end-to-end differentiable model for instance-wise feature selection. We jointly learn feature selection and prediction under a single objective without information leakage. Hide&Seek outperforms existing state-of-the-art models across a range of experiments and is fast to train. We achieve this by reformulating feature removal as a differentiable operation where instead of discretely removing features, we replace a proportion of each feature. Training is further stabilized via a parsimony-weight annealing framework.

We study the Random Quadratic Form (RQF) on a sphere in the presence of random Brownian forcing. We show that the forcing does not effectively change the law of the process but affects the synchronization properties of the system. While the RQF without forcing exhibits partial synchronization due to the intrinsic symmetries, the introduction of an arbitrarily small forcing results in long-term symmetry breaking and leads to full synchronization. In this work we focus on the small forcing regime and recover the multiscale behavior of the two-point process. We show that in the first stage the model converges to an anti-polar configuration due to the symmetries of the RQF and in the second stage the two clusters meet due to the symmetry breaking phenomenon. The model is motivated by continuous-time machine learning models such as Neural ODEs and continuous-time formulations of transformers. In particular, the results of this work explain the role of the bias and the scale of its initialization.

Hyperspectral reflectance spectroscopy enables non-destructive estimation of plant functional traits, yet current deep learning approaches process spectra as one-dimensional sequences, which limits how they capture long-range inter-band dependencies. We asked whether transforming 1D spectra into 2D image representations improves multi-trait prediction with convolutional neural networks (CNN). We compared nine transformations using EfficientNet-B0 on the GreenHyperSpectra dataset (7,897 labeled spectra, eight traits, 400-2450 nm), benchmarked against published 1D CNN results on the same split. Trained from scratch, the simplest transformation, a direct Reshape of the spectrum into a 2D grid, performed best ($R^2 = 0.684 \pm 0.001$) and improved on the state-of-the-art 1D baseline ($R^2 = 0.587$, $+0.097$). We then pretrained a 2D masked autoencoder (MAE-2D) on 139,000 unlabeled spectral images. Linear probing, which freezes the encoder and trains only a multilayer perceptron head, reached $R^2 = 0.646$ and exceeded every 1D self-supervised counterpart, including the fine-tuned MAE-1D ($R^2 = 0.641$). Under cross-dataset evaluation all models lost most of their accuracy and none beat the 1D baseline significantly. To identify which wavelengths drive each prediction, we applied Integrated Gradients and Grad-CAM and unfolded band importance back to the spectral axis. Protein ($r = 0.45$) and leaf water ($r = 0.33$) agreed with sensitivities simulated by the PROSAIL radiative-transfer model, while carotenoids ($r = 0.06$) and leaf area index ($r = -0.11$) did not, showing that the model reads established leaf chemistry for traits with sharp absorption features. The representational advantage of 2D spectral images, rather than architectural complexity or ImageNet pretraining, drives the gain over 1D approaches.

Ensembles of decision trees are well-established methods for data stream classification. In ensemble learning, Hoeffding Trees are widely adopted as base learners, performing periodic split attempts according to the Hoeffding bound. Recent studies, however, indicate that this standard splitting mechanism lacks adaptability, while adaptive trees that trigger splits in response to performance degradation have achieved superior results. In this paper, we identify limitations in the use of adaptive-splitting decision trees as ensemble base learners, showing that change detectors often fail to promote sufficient diversity within ensembles. To address this issue, we propose two novel decision tree models, termed Hoeffding Adaptive Splitting Trees. These models combine the periodic splitting strategy of Hoeffding Trees, which fosters ensemble diversity, with adaptive splitting mechanisms that employ change detection algorithms to identify performance decay and determine split points. Experimental results demonstrate that Hoeffding Adaptive Splitting Trees enhance ensemble performance and achieve state-of-the-art results across a comprehensive evaluation, including benchmark comparisons, computational cost analysis, and concept drift adaptation.

Automated detection of errors in clinical documentation is a promising application of large language models (LLMs), yet decisions to deploy such models rest on benchmarks that evaluate each clinical note in isolation. Error-detection benchmarks are typically constructed by injecting errors into notes, such that each erroneous note has a natural counterpart. Aggregate discriminative metrics (e.g., balanced accuracy or F1) do not exploit this structure. We show that this omission is consequential. In particular, evaluating 15 diverse LLMs on 4 standardized clinical error-detection test sets across 3 languages, we find that 13 of 15 models fall below the level of random pairwise discrimination, even while achieving F1 scores that standard practice would read as moderate. We also observe that the underlying bias patterns differ across languages: the same model can default to "no error" on one language and over-flag errors on another. To diagnose where discrimination breaks down, we further introduce a procedure to score the evidence models cite in their outputs. We find that while models consistently locate error-relevant content, they fail to produce the corresponding correct verdict on the clean counterpart. Finally, we show that F1 and pairwise accuracy are driven in opposite directions by the same underlying bias, so that ranking models by F1 may systematically promote the weakest discriminators. For safety-critical clinical NLP applications, we advocate for supplementing aggregate metrics with paired evaluations in benchmark reporting. Code and analysis scripts are available at https://github.com/healthylaife/paired-clinical-eval.

Repository-scale coding requires an agent to keep tests, imports, configuration, and migration rules consistent within a bounded context window. We model this as reconstructing a coupled-fact graph: at each edit, a required fact comes from recent context or parametric memory, and the facts covered by neither form coherence debt. We supply and withhold each channel and inject faults across seven models and five harnesses. As expected, no model completes a task on an unseen API with both channels empty, and putting the facts in the prompt restores success. When a rename defeats what models memorized about a real library, all seven fail in the same place, passing and missing the same tests. Availability decides the outcome and distance does not: withholding a fact costs exactly the work it supports, and a supplied fact works as well far from the edit as next to it. Harnesses pay unequal prices for it: configurations that all pass every test differ more than tenfold in tokens consumed because they rebuild the same content at different rates, and spending more recovers nothing when facts are withheld. A missing fact produces wrong work rather than absent work: an agent asked to act acts, fabricating the file or guessing the value, so instruments built on reads look for a hole already filled. How often it says it is blocked instead is a property of the model, from every trial to none. Availability does not settle every edit: where standard and code disagree, agents follow the standard even when it prescribes the worse code, so a stale convention file costs more than no file. Because parametric memory substitutes for reading, on SWE-bench, where models likely know the repositories, reads no longer predict success. Harnesses should keep the facts an edit depends on available when the agent writes, and check that availability against what the agent produces rather than what it reads.

An accurate estimation of the state of health (SOH) underpins safe and optimized use of the battery system. Although compelling, data-driven SOH estimation models typically require large amounts of high-quality labeled cycling data, while in practice such labels are often sparse in both quantity and coverage. Therefore, in this work, we propose a degradation-aligned self-supervised learning (SSL) framework based on a convolutional neural network-gated recurrent unit (CNN-GRU) model, which learns aging-consistent representations from unlabeled data through a cycle-order ranking objective as the pretext task for pretraining, thereby enabling robust SOH estimation after fine-tuning on sparsely labeled data. Test results showcase that the proposed ranking-based SSL approach proves to endow the pretrained model with degradation awareness from unlabeled data, and after fine-tuning the model can carry out accurate, robust SOH estimation, even when only an extremely limited amount of 1% of unevenly distributed labeled training data is available, where the MAE of 1.718% and RMSE of 2.329% can be achieved on the test cell. In addition, in-depth analyses are presented regarding the influences of label distribution and cross-cell robustness. We believe this work could shed new light on label-efficient SOH estimation of lithium-ion batteries, addressing a practical need in battery management.

Outliers can substantially distort Gaussian process regression (GPR) due to its conventional Gaussian observation likelihood, leading to inaccurate model learning and prediction. To address this limitation, this article introduces a generative GPR model that captures observation-specific contamination and adaptively mitigates the influence of outliers. Subsequently, a variational generalized expectation-maximization procedure is used to learn the latent variables and GPR model parameters. Experiments on synthetic and real datasets under different contamination settings demonstrate that the proposed method remains competitive with-and in several cases outperforms-robust GPR baselines in prediction accuracy. Moreover, the proposed method shares the cubic computational scaling of the compared GPR methods.

Accurate reconstruction of long-duration neural recordings is challenging because local field potentials (LFPs) are high-resolution, multichannel, transient, and variable across subjects. We present PCA-DMD, a scalable operator-theoretic framework that segments LFP recordings into overlapping windows, projects them into a compact PCA space, learns linear Koopman evolution in the latent space, and reconstructs continuous signals through inverse projection and overlap-add aggregation. On 200,000-sample hippocampal recordings, PCA-DMD outperformed Classical DMD, SpDMD, MrDMD, and HODMD, achieving KLD=0.0761 and HD=0.0847. In all-pair cross-subject zero-shot generalization at 300,000 samples, correlations were 0.9504-0.9800, with HD=0.0010-0.0072 and KLD=0.0005-0.0022, without target-subject fine-tuning. The prediction showed close one-step agreement on temporally held-out LFP segments across the unseen interval and multiple channels. Scalability analysis from 400,000 to 900,000 samples showed stable zero-shot reconstruction, with mean correlation remaining about 0.965-0.968 while computational cost increased predictably. External validation on an independent 93-channel Allen Neuropixels recording yielded mean and median channel-wise correlations of 0.7427 and 0.7990, respectively. Koopman spectral and mode analyses revealed dominant eigenvalues concentrated near the unit circle. PCA-DMD therefore provides an interpretable, generalizable, and computationally scalable framework for reconstructing high-dimensional neural dynamics.

This cumulative habilitation thesis studies probabilistic circuits (PCs) as a powerful and tractable framework for reasoning and learning under uncertainty in artificial intelligence (AI). It first advocates for probability as a core language for AI, emphasizing its connections to logic and information theory; the conceptual simplicity of probabilistic reasoning—based primarily on the sum and product rules; the parallels between probabilistic inference and human cognition; and the role of probability in optimal decision making. However, probability also faces significant computational challenges, as probabilistic inference is NP-hard in almost all probabilistic models. PCs address these challenges through structural constraints that ensure exact computation of a wide range of inference queries in polynomial time, such as marginals, conditionals, most probable explanations, expectations, and more advanced inference tasks. This thesis synthesizes a decade of research across foundations, algorithmic developments, and empirical validation of PCs. Key contributions highlighted in this work are foundational theory of PCs, Bayesian approaches for learning PCs, scalable implementations and integration with deep learning, hybrid models that combine PCs with intractable models, and connections with symbolic machine learning paradigms. This is the first part of my Habilitation Thesis. The second part is omitted, as it comprises the cumulative part of the thesis and has been published at various venues (see Chapter 5).

Most super-resolution models learn from paired data by supervising only the final high-resolution output. This provides little control over how the prediction should evolve between the downsampled observation and its fine target. We introduce GalerkinFlow, an equation-agnostic framework that turns each coarse–fine pair into supervision along an entire reconstruction path. At a random sample of intermediate states on the reconstruction path, the model predicts the coarse-to-fine residual velocity and uses coarse-anchor point to define a pseudo-endpoint. We show that the reconstruction loss of this pseudo-endpoint is exactly related to the intermediate velocity loss through a known time-dependent weight. Consequently, every intermediate state contributes supervision toward the same fine target, rather than serving only as an internal step toward an endpoint loss. Because intermediate states already reveal part of the missing fine-scale structure, we additionally supervise the coarse endpoint used during one-step inference. A finite-difference objective further constrains local spatial variation. GalerkinFlow combines convolutional features with scale-conditioned Galerkin operator mixing and requires no governing equation or physical metadata. It achieves the lowest raw-space errors among the evaluated equation-agnostic baselines on Navier–Stokes and Darcy Flow, while remaining competitive on DIV2K.

Continuous wrist-worn hand sensing for gesture interfaces and motor symptom monitoring needs an always-on front end that fits inside a coin-cell power budget while pairing a micro-electro-mechanical-systems (MEMS) inertial measurement unit (IMU) with a 60 GHz frequency-modulated continuous-wave (FMCW) radar to stay robust under occlusion and on-body drift. We present a design study of such a wristband front end in which classifier wake-up gating, mmWave versus IMU routing, and innovation-based EKF measurement reweighting share a single on-chip residual generator. The shared generator occupies 14.4 KB of program memory and 278 B of state and runs at 110K multiply-accumulates (MACs) per frame on an Ambiq Apollo4 Blue Plus class edge microcontroller unit (MCU). Across four public sensor data corpora (IPN Hand, SHREC 2021, MiliPoint 60 GHz FMCW radar, EAT-Radar) the front end reaches detection probability $P_D = 0.72/0.80$ at a 1% false-alarm rate, sustains a 47% classifier invocation energy reduction at 90% gesture detection recall, and lowers pose tracking root-mean-square error by $4.6\times$ under measurement bias drift relative to an adaptive Kalman with $R$-inflation baseline. Measured silicon power and on-body capture are deferred to follow-on hardware; the contribution here is a design study.

Objective assessment of learning remains a fundamental challenge in education. Electroencephalography (EEG) provides a direct, non-invasive window into the neural correlates of knowledge acquisition, including cognitive familiarity. This study benchmarks fifteen machine learning (ML) and deep learning (DL) models for EEG-based familiarity prediction across two cognitive domains: faces (factual knowledge) and mathematical equations (conceptual knowledge). Using continuous EEG data from 23 participants, we extract spectral features (Power Spectral Density) across six frequency bands. We show that while standard stratified cross-validation yields artificially high classification performance (up to 0.9853 F1-score using CNN) due to temporal leakage across neighboring epochs, a rigorous trial-independent validation (Group K-Fold) drops the peak performance to 0.6038 F1-score (using CNN), which is still statistically significant above the 25% chance level. This highlights the critical necessity of trial-independent evaluation to avoid overestimating model generalizability. Furthermore, feature importance and SHAP analysis reveal that temporal and frontal Gamma and Beta oscillations are the most critical biomarkers for familiarity. This work establishes a realistic benchmark for EEG-based cognitive monitoring in educational technologies.

Spatially resolved EEDFs/IEDFs provide essential kinetic information about low-temperature plasmas (LTPs) and play a central role in determining transport, chemical reaction rates, and plasma surface interactions. While kinetic simulations directly resolve these distributions, experimental measurements remain challenging and are often invasive, spatially limited, or require assumptions regarding the distribution shape such as a Maxwellian. However, several macroscopic plasma observables can be measured non-invasively using advanced diagnostic techniques, providing spatially resolved information about the plasma state. An important inverse problem is therefore whether readily measurable macroscopic plasma quantities contain sufficient information to reconstruct the underlying kinetic state. In this work, we investigate this problem by learning a nonlinear mapping from spatially resolved macroscopic plasma observables to the corresponding spatially resolved EEDFs/IEDFs using a deep learning framework. Paired datasets comprising 2D macroscopic observables and spatially resolved EDFs are generated using 2D-3V PIC-MCC simulations. Three representative learning paradigms, a U-Net, a FNO, and a MeshGraphNet, are employed in this study to learn this inverse mapping. The predicted EDFs reproduce both bulk plasma and sheath characteristics with good agreement to the PIC-MCC reference data, with the FNO providing the best overall performance. Beyond conventional metrics, physics-based validation demonstrates that the reconstructed EDFs accurately recover the corresponding density and temperature, and rate coefficients. These results demonstrate that macroscopic plasma observables encode sufficient information to infer important kinetic properties in LTPs, providing a potential foundation for surrogate kinetic modeling and next-generation plasma diagnostics.

We propose a two-stage large language model (LLM) framework for zero-shot detection of insider threats and advanced persistent threats (APTs) from heterogeneous security logs. The framework models user activity as chronological timelines and incorporates retrieval-augmented generation (RAG) to provide personalised behavioural context from each user's historical activity. Rather than performing end-to-end classification directly from raw logs, it first generates structured, interpretable sets of threat-specific risk indicators, which are then classified jointly across temporal sequences to capture attack patterns spanning multiple windows.The framework is evaluated on two benchmark datasets, CERT r5.2 for insider threat detection and PicoDomain for APT detection, using four combinations of two open-weight LLMs under both retrieval and non-retrieval settings. All configurations outperform the previous state-of-the-art LLM-based framework (GABM), with the best configuration improving the F1-score by 11.40 percentage points on CERT r5.2 and 31.50 percentage points on PicoDomain. Results further show that retrieval mainly benefits weaker LLMs by generating more discriminative risk indicators, whereas stronger models achieve comparable performance without retrieved context. The most effective assignment of LLMs to the two stages depends on the dataset. These findings show that the quality of the generated risk indicators is the main driver of zero-shot cyber threat detection performance.

Dataset alignment is a central step in data analysis across science and engineering, where the goal is to match observations between datasets. Entropic Optimal Transport (EOT) offers a computationally tractable framework for this task by encoding cross-dataset affinities in a transport plan. However, when two datasets are sampled from geometrically similar low-dimensional structures with substantially different sampling densities, the EOT plan may match points by relative sampling density rather than geometric proximity, yielding geometrically misleading correspondences. To address this issue, we propose a density-reweighted EOT framework in which the influence of sampling density on the transport plan can be discounted to a desired degree, ranging from standard EOT to alignment driven purely by underlying geometry. Under suitable regularity conditions, we establish convergence of the reweighted EOT plan to a family of population-level plans whose dependence on sampling density is made explicit. Through simulations, we show that our approach recovers geometrically faithful correspondences, improving over related EOT-based frameworks when datasets exhibit substantial sampling density disparity.