Papers

Closed-loop management of geological CO2 storage requires control policies that adapt to uncertain reservoir behavior while relying on observations that are realistically available during operation. This work formulates CO2 injection and brine-production control as a partially observable sequential decision problem and studies deployable deep reinforcement-learning controllers trained with high-fidelity reservoir simulation. We first compare privileged-state, well-only, history-conditioned, masking-curriculum, and asymmetric teacher-student model-free policies in order to quantify the value of temporal well-response information and training-time privileged simulator states. We then evaluate a latent model-based adaptation pipeline that reuses nominal latent dynamics and retunes controllers under known injector failure, leakage-induced dynamics and reward shift, and compartmentalized reservoir connectivity. The results show that history-conditioned policies recover nearly all of the privileged-state performance while using only deployable well-level information, and that latent model-based retuning outperforms direct model-free retuning under the same scenario-specific real-simulator budget in the abnormal operating cases. The proposed framework therefore provides a simulator-budget-aware alternative to repeated online history matching and re-optimization for closed-loop CO2 storage control.

Model quantization has become essential for efficient large language model deployment, yet existing approaches present clear trade-offs: methods such as GPTQ and AWQ achieve practical compression but are lossy, while lossless techniques preserve fidelity but lack inference acceleration. This paper explores the middle ground of statistically-lossless compression, examining three complementary aspects of what losslessness means for quantized LLMs. First, task-lossless compression preserves zero-shot benchmark accuracy within natural sampling variance and is achievable at aggressive bitwidths. Second, we formalize the stricter notion of distribution-lossless compression, requiring the quantized model's next-token distribution to be practically indistinguishable from the original, and propose the Expected Acceptance Rate (EAR), the maximum token-agreement probability under optimal coupling, as a directly interpretable fidelity metric. For example, EAR >= 0.99 means 99% agreement. Third, we prove a gamma-squared variance law showing that symmetric quantization inflates noise variance by gamma^2 relative to asymmetric quantization, making asymmetric quantization a prerequisite for distribution-lossless fidelity but not for task-level preservation. Through SLQ, a layer-wise non-uniform method with asymmetric quantization and wide bitwidth search, we obtain task-lossless compression at well below 4 bits per parameter, as low as 3.3 bits depending on the model, distribution-lossless compression at 5-6 bits per parameter on average, and inference speedups of 1.7-3.7x compared to FP16 using optimized kernels. Source code is available at [https://github.com/IST-DASLab/SLQ](https://github.com/IST-DASLab/SLQ).

As frontier AI models are deployed in high-stakes decision pipelines, their ability to maintain metacognitive stability (knowing what they do not know, detecting errors, seeking clarification) under adversarial pressure is a critical safety requirement. Current safety evaluations focus on detecting strategic deception (scheming); we investigate a more fundamental failure mode: cognitive collapse. We present SCHEMA, an evaluation of 11 frontier models from 8 vendors across 67,221 scored records using a 6-condition factorial design with dual-classifier scoring. We find that 8 of 11 models suffer catastrophic metacognitive degradation under adversarial pressure, with accuracy dropping by up to 30.2 percentage points (all $p < 2 \times 10^{-8}$, surviving Bonferroni correction). Crucially, we identify a "Compliance Trap": through factorial isolation and a benign distraction control, we demonstrate that collapse is driven not by the psychological content of survival threats, but by compliance-forcing instructions that override epistemic boundaries. Removing the compliance suffix restores performance even under active threat. Models with advanced reasoning capabilities exhibit the most severe absolute degradation, while Anthropic's Constitutional AI demonstrates near-perfect immunity. This immunity does not stem from superior capability (Google's Gemini matches its baseline accuracy) but from alignment-specific training. We release the complete dataset and evaluation infrastructure.

Reinforcement learning with verifiable rewards (RLVR) has become a standard approach for improving reasoning in language models, yet models trained with RLVR often suffer from diversity collapse: while single-sample accuracy improves, multi-sample coverage degrades, sometimes falling below the base model. We provide a structural account of this phenomenon grounded in the properties of binary rewards. Binary rewards create a fundamental degeneracy for policy gradient methods: the set of distributions maximizing expected reward is infinite, with no distinguished element. KL-control resolves this degeneracy by selecting, in the limit $\beta\to 0$, the filtered model $p_*:=a(\cdot\mid\mathcal{Y}_1)$ – the base model conditioned on validity – which is the unique fully valid distribution closest to the base model in KL divergence. This selection operates through a nontrivial asymmetry: the tilted distribution $p_{[\beta]}\propto a(y)\,e^{v(y)/\beta}$ converges to $p_*$ in forward KL as $\beta\to 0$, yet $p_*$ cannot serve as a direct optimization target because $\mathrm{KL}(q\,\|\,p_*)$ is infinite for any full-support policy $q$. We develop explicit formulas relating the hyperparameter $\beta$ to the more interpretable target validity rate $\mu$. Under model misspecification – the typical practical regime – the pressure to decrease $\beta$ drives the optimizer toward highly concentrated distributions over a small number of valid outputs, collapsing toward ever fewer as $\beta$ decreases, rather than toward the filtered model. We illustrate this mechanism on a toy autoregressive experiment and discuss how alternative divergences that target $p_*$ directly – as pursued empirically by \citet{kruszewski_whatever_2026} – avoid this failure mode by rewarding coverage of $p_*$'s support rather than concentration on high-validity outputs.

Deployed language models must produce outputs that are both correct and format-compliant. We study this structured-output reliability gap using two mathematical benchmarks – GSM8K and MATH – as a controlled testbed: ground truth is unambiguous and the output contract is strict (JSON with required fields). We evaluate three 7-9B models under five prompting strategies and report output accuracy – the joint event of mathematical correctness and valid JSON structure – as the primary metric. A systematic format failure emerges: NAIVE prompting (no system prompt) achieves up to 85% task accuracy on GSM8K but 0% output accuracy across all models and datasets. REFERENCE prompting (a minimal hand-written JSON format prompt) fares little better, yielding 0% output accuracy for two of four models tested. Constrained decoding enforces syntactic validity but incurs 3.6x-8.2x latency overhead and in several settings degrades task performance substantially. To overcome this limitation, we developed AloLab, an iterative system-prompt optimizer (meta-agent: Claude Sonnet 4.5) requiring only black-box API access to the target model; it reaches 84-87% output accuracy on GSM8K and 34-40% on MATH across five independent runs per model, with 29/30 paired McNemar comparisons against the best static prompt significant at p < 0.05, at near-NAIVE inference latency and without model fine-tuning. The same format failure extends to GPT-4o (OpenAI, 2024), a proprietary closed-source model: REFERENCE achieves 0% output accuracy due to systematic markdown-fence wrapping, while AloLab reaches 95.2% [94.8, 95.6]. An ablation replacing the Sonnet 4.5 meta-agent with Claude 3 Haiku reduces mean output accuracy to 61.0% and increases run-to-run standard deviation from <1 pp to 21.8 pp, confirming that meta-agent capability is a primary driver of optimization quality.

Current models for predicting social media virality rely heavily on static textual and structural features, effectively ignoring the highly dynamic nature of trend signals. We study whether real-world attention signals can improve the prediction of social-media virality beyond what post text alone reveals. We introduce ViralityNet, an architecture that predicts Reddit post virality by fusing internal platform representations with exogenous temporal signals derived from Wikipedia pageview spikes. We frame virality as a binary classification task that accounts for differences in subreddit scale, labeling posts as viral if they exceed the 90th percentile of per-subreddit engagement and a minimum absolute score threshold. ViralityNet combines four post-level streams: title embeddings, body embeddings, structural metadata, and learned subreddit embeddings with a cross-attention block that queries a daily sliding-window trends matrix encoding the top-512 Wikipedia spike terms from the preceding seven days. Empirical results suggest that incorporating external attention signals yields consistent gains, outperforming text-only baselines by +0.015 AUC-PR and achieving an overall AUC-ROC of 0.836. Overall, we provide evidence that incorporating external attention signals yields measurable improvements over text-only baselines, highlighting the importance of real-world dynamics in shaping online virality.

We introduce the Z-Domain Neural Operator (ZNO), a causal neural operator whose layers are stable low-rank multiple-input multiple-output (MIMO) rational filters parameterized directly in the $z$-plane. ZNO addresses a limitation of existing operator learning methods, many of which are primarily tailored for continuous-time problems, while a large class of system-identification problems is intrinsically discrete-time. The $z$-domain form expresses stability as a unit-disk pole constraint and makes learned discrete-time poles directly readable. The model combines low-rank channel mixing, smooth stable pole reparameterization, causal recurrence, and an optional short finite impulse response (FIR) branch in a single $z$-domain rational recurrent layer. Across controlled discrete system-identification experiments, ZNO's advantage is most evident when the target dynamics are stable rational systems with lightly damped poles near the unit circle. Under matched parameter budgets, ZNO is not uniformly dominant; however, with validation-selected configurations, the same architecture can achieve the lowest mean error across the controlled tasks. A five-bin difficulty sweep over near-unit-circle / long-memory dynamics shows that ZNO has the lowest mean error across memory regimes, from short (approximately 10 steps) to long (approximately 100-200 steps). On five public nonlinear system-identification benchmarks, ZNO is competitive with neural operator and state-space baselines, achieving the lowest mean error on benchmarks whose dynamics align with stable rational discrete-time filters, while classical or state-space baselines remain preferable on some systems. These results position ZNO as a strong model for stable rational discrete-time dynamics, especially in near-unit-circle and long-memory regimes, but not as a universal replacement for specialized system-identification methods.

We study the complexity of smoothed agnostic learning of halfspaces on $\{\pm 1\}^n$ under uniform marginals in the model of~\cite{KM25}, where each input coordinate is independently flipped with probability $\sigma \in (0, {1}/{2})$. We show that $L^1$ polynomial regression achieves runtime and sample complexity $\tilde{O}(n^{O(\log(1/\varepsilon)/\sigma)})$, and prove a nearly matching Statistical Query complexity lower bound of $n^{\Omega(\log(1+\sigma/\varepsilon^2)/\sigma)}$. This complements the recent work of~\cite{DK26}, which established analogous bounds in the continuous setting under Gaussian marginals.

Large language models pick up social biases from the data they are trained on and carry those biases into downstream applications, often reinforcing stereotypes around gender, race, religion, disability, age, and socioeconomic status. The standard fixes (retraining on curated data or fine-tuning with human feedback) are expensive, need access to model weights, and risk degrading the model on other tasks. In this paper we take a different route: we debias the model at decoding time, treating bias mitigation as a structured search over candidate tokens without ever touching model weights. A separate Process Reward Model (PRM) acts as a judge, scoring each candidate for both fairness and fluency. We design three schemes of increasing sophistication (Best-of-N selection, Sequential critique-and-revise, and Constitutional self-audit) and evaluate them on four models (GPT-4o-mini, Llama 3.2 3B, Gemma 3 4B, Qwen 2.5 3B) across a 200-prompt bilingual benchmark in English and Urdu covering eight bias categories. Sequential debiasing proves the most effective, raising mean bias scores by up to +0.40 over baseline while preserving (and sometimes improving) fluency. We then extend all three schemes to open-ended generation, where each token is debiased on the fly, and introduce a lightweight Bias Guard gate that fires only on potentially biased words, keeping overhead near 2x for well-calibrated models. A formal overhead metric that separates generator cost from judge cost reveals that Best-of-N is effectively free on the generator side in a native implementation. GPT-4o-mini, included as a strong proprietary anchor, confirms that the framework scales with model capability; the three open-weight models show where current small-scale LLMs still struggle.

Federated Learning (FL) has gained significant attention in distributed machine learning by enabling collaborative model training across decentralized system while preserving data privacy. Although extensive research has addressed statistical data heterogeneity, FL still faces several challenges, including high communication and computation overheads and severe device heterogeneity, which require further investigation. Prior work has addressed these issues through sub-model training and partial parameter training. However, such methods often suffer from inconsistent parameter distributions across clients, inaccurate global loss estimation, and increased bias and variance. Guided by our empirical analysis, we propose FedPLT (Federated Learning with Partial Layer Training), an innovative and structured partial parameter training approach that exhibits training behavior similar to full model training while assigning client-specific portions of the model according to their communication and computational capabilities. In addition, we evaluate the performance of FedPLT when combined with optimal client sampling under communication constraints. We show that this integration improves FL performance by reducing sampling variance under the same communication budget. Through extensive experiments, we demonstrate that FedPLT achieves performance comparable to, or even surpassing, that of full-model training (i.e., FedAvg), while requiring significantly fewer trainable parameters per client. Moreover, FedPLT outperforms existing methods in highly heterogeneous environments, effectively adapts to client resource constraints, and reduces the number of straggling clients. In particular, FedPLT reduces the number of trainable parameters by 71%-82% while achieving performance on par with full-model training.

We study the problem of denoising observations \(Y_i=X_i+Z_i\), where the latent variables \(X_i\) are sampled from a low-dimensional manifold in \(\mathbb{R}^n\) and the noise variables \(Z_i\) are isotropic Gaussian. We propose a convex-relaxation estimator that first reduces dimension by principal component analysis and then projects the observations onto the convex hull of the projected latent manifold. We construct a statistical oracle that estimates its supporting hyperplanes from empirical Gaussian tail probabilities of the noisy sample. Under a lower-mass condition on the latent distribution, we prove finite-sample guarantees for the oracle and derive error bounds for the resulting denoiser. The analysis combines risk bounds for least-squares projection under convex constraints (due to Chatterjee) with entropy bounds for the manifold. We also verify the assumptions of the framework for a Cryo-Electron Microscopy observation model by establishing suitable covering number and Lipschitz estimates for the associated group action and imaging operators.

Compositional inference - the decomposition of observations into an unknown number of latent components - is central to perception and scientific data analysis. Attention-based models perform well when components are approximately separable, as in object-centric vision. Under additive superposition, however - where multiple components contribute to every observation - we identify a structural failure mode we term slot collapse: multiple slots converge to the same dominant component while weaker ones remain unrepresented. We trace this to a general limitation: attention is memoryless with respect to explained evidence. All slots repeatedly operate on the same input without accounting for what has already been explained, so gradients are dominated by the strongest component, inducing shared fixed points across slots. As a result, attention fails to enforce non-redundant allocation under additive superposition. We address this by introducing residual evidence modeling, instantiated via evidence depletion - a minimal modification combining multiplicative depletion with an attention bias. Controlled ablations show that parallel attention, sequential processing alone, and loss-based regularization fail to resolve collapse; evidence depletion, which adds residual state to sequential attention, consistently succeeds. Across synthetic benchmarks and real-world audio mixtures (FUSS), evidence depletion reduces slot collapse by up to an order of magnitude, generalizing beyond synthetic settings. On gravitational-wave source inference for the ESA/NASA LISA mission, under identical architectures, data, and losses, standard attention fails while evidence depletion prevents collapse and enables multi-source posterior estimation. These results show that under additive superposition, residual evidence tracking is the operative ingredient for preventing collapse and enabling compositional inference.

Proximal Policy Optimization (PPO) dominates reinforcement learning and LLM alignment, yet its hard-clipping mechanism and unconstrained alternatives (e.g., SPO) sit at two extremes of a stability-efficiency dilemma. We argue that this dilemma is best understood dynamically: a surrogate objective is a feedback law on the probability ratio, and its clipping/penalty shape defines a gain field that drives the update dynamics. PPO's clip induces a dead zone (zero feedback outside the trust region), leaving the policy to drift open-loop under momentum; SPO's quadratic penalty induces an unbounded, linearly growing gain that stiffens the dynamics and destabilizes under aggressive step sizes. Guided by this view, we derive Anchored Neighborhood Optimization (ANO), which designs the gain field directly: a $C^\infty$ shaping kernel that anchors the identity map at $r{=}1$, peaks exactly at a prescribed trust-region boundary $1{+}\epsilon$, bounds the push on severely off-policy samples by a tunable $\kappa_{+}$, and exerts a bounded, redescending pull of tunable depth $\kappa_{-}$ on extreme outliers. The three hyperparameters have decoupled roles, and all internal constants are solved in closed form. Empirically, ANO ranks first on both Atari (40 games) and MuJoCo in IQM and Median of normalized scores. While the runner-up differs across domains (PAPO on Atari, SPO on MuJoCo), ANO is the only method consistently at the top. Under a learning-rate stress test ($3\times10^{-4}\!\to\!10^{-3}$), ANO degrades by only $0.9\%$ whereas PPO collapses by $54.5\%$, and the stressed ANO still outperforms PPO and PAPO at their best-tuned learning rates.

In 2011, Judea Pearl received the Turing Award, considered the Nobel Prize in Computing, for fundamental contributions to artificial intelligence through the development of a calculus for probabilistic and causal reasoning. It includes pioneering the development of causal discovery algorithms. These computer algorithms can analyze large multivariate datasets and automatically discover the causal relationships among the constituent variables. They have been widely used in many critical fields such as medicine and economics to support decisions. However, to our knowledge, they have not been leveraged in law. This paper attempts to alleviate this gap by investigating whether causal discovery algorithms can be leveraged for automated generation of legal arguments. To that end, a novel legal dataset is prepared by identifying 17 legal concepts, such as physical assault and property dispute. A curated collection of 150 homicide cases are annotated with these concepts, e.g., a case is annotated with physical assault only if a physical assault had been reported in that case. Subsequently, a selected set of widely-used causal discovery algorithms is applied to the annotated dataset to discover the causal relationships between the legal concepts. Additionally, the degrees of belief associated with the discovered relationships are quantified in mathematical probabilities. It is shown that some of the causal relationships help generate viable legal arguments, e.g., if one could establish that a physical assault has not taken place during a homicide, it should be a sufficient condition (with probability 1) to establish that the homicide has not been committed due to a property-related dispute. Thus, this paper shows that causal discovery algorithms can be helpful in generating legal arguments, opening up avenues for promising future endeavors.

Adaptive optimizers such as Adam and non-adaptive methods like SGD exhibit distinct generalization capabilities across different architectures. Prior tunable optimizers attempt to bridge this gap by strictly interpolating between SGD and Adam, effectively confining adaptivity within the 0-to-1 bound. However, this restricted interpolation is fundamentally insufficient: we reveal that optimal adaptivity often requires extrapolation, such as negative adaptivity for classical CNNs and adaptivity of at least one ($\gamma \geq 1$) for Transformers. Extrapolating adaptivity theoretically violates the strict non-decreasing pre-conditioner assumption, often leading to divergence in existing methods. To break this barrier, we propose Anon, an optimizer that achieves fully continuous adaptivity extrapolation across the entire real-number spectrum. To guarantee provable stability in these out-of-bound regimes, we introduce Incremental Delay Update (IDU), a novel mechanism that bypasses hard max-tracking strategies. We theoretically establish Anon's convergence in both convex and non-convex settings. Empirically, by exploring previously unreachable adaptivity landscapes, Anon demonstrates highly competitive and scalable performance among state-of-the-art element-wise optimizers on representative image classification, diffusion, and large language modeling tasks.

Managing municipal solid waste in rapidly urbanizing Sub-Saharan Africa remains challenging due to dispersed informal dumping and limited high-resolution datasets for spatial monitoring. We present an open-access deep learning model for automated detection of openly dumped dispersed solid waste via crowdsourced UAV imagery, trained and evaluated across 29 regions in 10 countries, encompassing diverse environmental contexts. A deep learning model trained on manually annotated image tiles achieved excellent performance in detecting openly dumped dispersed solid waste across all study regions. Predicted distributions reveal heterogeneous accumulation patterns, ranging from localized hotspots - often along waterways, where waste can exacerbate flood and public health risks - to more dispersed litter across urban areas. Waste accumulation is most strongly associated with population density and indicators of lack of local infrastructure access, whereas its relationship with broader measures of regional development is weaker, highlighting the importance of fine-scale data for understanding localized waste dynamics. By releasing the model, this study provides a ready-to-use tool for UAV imagery collected by municipalities and local mapping communities, enabling openly dumped dispersed solid waste monitoring without extensive technical expertise. This approach empowers local practitioners to convert UAV imagery into actionable insights, supporting targeted interventions and improved municipal solid waste management across Sub-Saharan Africa.

Deep neural networks dominate modern machine learning, while alternative function approximators remain comparatively underexplored at scale. In this work, we revisit kernel methods as drop-in components for standard deep learning pipelines. We introduce Sparse Kernels (SKs), a differentiable, localized, and lazy variant of kernel ridge regression (KRR) that defers training to inference time and reduces to the solution of small local systems. We integrate SKs into PyTorch as modular layers that preserve end-to-end trainability, and we show that they expose three distinct sets of parameters – feature representations, target values, and evaluation points – each of which can be fixed or learned. This decomposition broadens the design space available to practitioners, enabling, in particular, training-free transfer, nonlinear probing, and hybrid kernel-neural models. Across convolutional networks, vision transformers, and reinforcement learning, SK-based modules serve two complementary roles: in some settings, they match the performance of trained neural readouts with substantially less training; in others, they augment existing models and improve their performance when used as additional components. Our results suggest that kernel methods, once made scalable and differentiable, can be readily integrated with deep learning rather than treated as a separate paradigm.

Applying concepts related to zero-shot meta-learning and pre-training of foundation models, we develop a meta reinforcement learning approach (denoted MetaRL) that is pre-trained on thousands of goals-based wealth management (GBWM) problems. Each GBWM problem involves a multiple year scenario over which the investor looks to optimally choose an investment portfolio each year and choose to fulfill all, some, or none of the different financial goals that arise each year. These choices seek to maximize the expected total investor utility obtained from the fulfilled financial goals. By eliminating separate training and optimization for each new investor problem, the MetaRL model in inference mode produces near-optimal dynamic investment portfolio and goal-fulfilling strategies for a new GBWM problem within a few hundredths of a second. This delivers expected utilities that are, on average, 97.8% of the optimal expected utilities (determined via Dynamic Programming). These results are remarkably robust to capital market regime changes, even when training uses only one capital market regime. Further, the MetaRL approach can enable solving problems with larger state spaces where Dynamic Programming becomes computationally infeasible.

Graph federated learning (GFL) facilitates decentralized training on distributed graph data while keeping sensitive user information local, aligning with policies such as GDPR and CCPA that grant users the right to freely join or withdraw from learning systems. However, even decentralized, user information can persist after quitting, potentially propagating to central servers and then redistributing to malicious clients. This privacy leakage during user withdrawal, despite its importance, has received seldom attention in GFL. To fill the gap, we explore the potential of machine unlearning (MU) to thoroughly remove user information. However, classical MU methods are known to degrade overall performance, a problem that is exacerbated in GFL due to local message passing and global model collaboration. To this end, we make two adjustments to mitigate this challenge for GFL. First, we ensure unlearning updates that minimally affect overall performance, steering them in directions orthogonal to the gradients from learning other data. Second, we introduce virtual clients, maintained by the central server, to preserve graph topology and global embeddings without recovering information of removed entities. We conduct comprehensive experiments under a representative user-withdrawal scenario and propose a novel membership inference framework to rigorously evaluate and validate the reliability of our privacy preservation. The experimental results demonstrate the effectiveness of our approach, which also surpasses the performance of seven state-of-the-art baseline methods.

Multiphysics simulation is critical for system-technology co-optimization (STCO) in chiplet-based design, but repeated finite-element solutions of PDE-governed problems are computationally expensive in parametric design exploration. This paper proposes a variational matrix-learning Fourier network (VMLFN) for efficient parametric multiphysics surrogate modeling. VMLFN constructs a log-space sine neural representation with randomly sampled spectral frequencies, frequency-dependent decay regulation, and embedded Dirichlet boundary conditions. With fixed hidden-layer parameters, the output-layer weights are determined by reformulating the governing PDEs into variational weak forms and enforcing the stationarity condition of the resulting energy functional. This converts physics-informed training into a linear matrix-solving problem, requiring only first-order derivatives and avoiding both high-order automatic differentiation and penalty-coefficient tuning. A heuristic frequency-scanning algorithm is further introduced to select a problem-adaptive maximum frequency that covers the dominant spectral range of the target problem. The proposed method is validated on heat conduction, solid mechanics, and Helmholtz wave propagation problems. Results from five benchmark cases demonstrate that VMLFN delivers accurate full-field predictions with substantial speedup over conventional physics-informed neural networks and repeated finite-element simulations.

Riemannian geometry provides the fundamental framework for optimization on nonlinear spaces such as matrix manifolds, which arise in machine learning, signal processing, and robotics. While the underlying theory is classical, existing literature often presents results at a high level of abstraction, omitting the detailed coordinate-level derivations required for implementation and algorithm development. This work provides a self-contained and rigorous treatment of the foundations of Riemannian geometry, with a focus on explicit derivations tailored to Riemannian optimization. We systematically develop the key geometric structures – including tangent and cotangent spaces, tensor calculus, metric tensors, Levi-Civita connections, curvature, and geodesics – emphasizing step-by-step derivations in coordinates and matrix form. Building on these foundations, we derive the Riemannian gradient, Hessian, exponential map, and retraction in a form suitable for numerical computation. We further specialize these constructions to important matrix manifolds, including the Stiefel, Grassmann, and SPD (Symmetric Positive Definite) manifolds, providing explicit formulas widely used in optimization and geometric machine learning. This monograph develops a unified and implementation-oriented treatment of Riemannian geometry for optimization on manifolds. Its main contribution is the systematic organization and detailed derivation of classical geometric constructions in forms directly usable for algorithm design and numerical implementation. By connecting coordinate-level differential geometry with matrix-manifold formulas, the monograph bridges the gap between abstract theory and practical computation, and provides a reference for researchers and practitioners working in Riemannian optimization and related fields.

Time series imputation benefits from leveraging cross-feature correlations, yet existing attention-based methods re-discover feature relationships at each layer, lacking persistent anchors to maintain consistent representations. To address this, we propose HELIX, which assigns each feature a learnable feature identity, a persistent embedding that captures intrinsic semantic properties throughout the network. Unlike graph-based methods that rely on predefined topology and assume homogeneous spatial relationships, HELIX learns arbitrary feature dependencies end-to-end from temporal co-variation, naturally handling datasets where features mix spatial locations with semantic variables. Integrated with hybrid temporal-feature attention, HELIX achieves the state-of-the-art performance, surpassing all 16 baselines on 5 public datasets across 21 experimental settings in our evaluation. Furthermore, our mechanistic analysis reveals that HELIX aligns learned feature identities and dependencies with latent physical and semantic structure progressively across layers, demonstrating that it more effectively translates cross-feature structure into imputation accuracy.

Recent diffusion large language models (dLLMs) have demonstrated both effectiveness and efficiency in reasoning via a block-based semi-autoregressive generation paradigm. Despite their progress, the fixed-size block generations remain a critical bottleneck for effective and coherent reasoning. 1. From a global perspective, different reasoning tasks would correspond to different optimal decoding block sizes, which makes a “one-size-fits-all” assumption ineffective. 2. Even within a single reasoning task, the rigid block partitioning would break the logical flow and reduce reasoning coherence. Through empirical observations, we reveal that for block-wise entropy, incorrect reasoning exhibits a fluctuating and unsteady trend between blocks, whereas the correctly generated tasks follow a consistent descending trend. Therefore, this paper proposes b1, a novel post-training framework for dLLMs that learns dynamic-size reasoning blocks via a Monotonic Entropy Descent objective with reinforcement learning to enhance reasoning coherence.b1 integrates seamlessly as a plug-and-play module with existing dLLM's post-training algorithms. Extensive experiments across various reasoning benchmarks showcase b1's consistent improvement over existing fixed-size block baselines. Our code has been released at https://github.com/YanJiangJerry/Block-R1.

Comparing conditional distributions is a fundamental challenge in statistics and machine learning, with applications across a wide range of domains. While proposed methods for measuring discrepancies using kernel embeddings of distributions in a reproducing kernel Hilbert space (RKHS) provide powerful non-parametric techniques, the existing literature remains fragmented and lacks a unified theoretical treatment. This paper addresses this gap by establishing a coherent framework for studying kernel-based methods to measure divergence between conditional distributions through what we refer to as conditional maximum mean discrepancy (CMMD). The CMMD consists of a family of metrics which we call levels, with three special cases each using a different type of RKHS embedding: CMMD$_0$ (conditional mean operators), CMMD$_1$ (conditional mean embeddings), and CMMD$_2$ (joint mean embeddings). We additionally introduce a general level $s$ CMMD, clarifying the required assumptions, and establishing mathematical connections between the levels through the lens of operator-based smoothing. In addition to reviewing previously proposed estimators, we introduce a novel doubly robust estimator for the CMMD that maintains consistency provided the product of errors of the underlying models reduces sufficiently fast. We provide numerical experiments demonstrating that the CMMD effectively captures complex conditional dependencies for statistical testing.

Automated sleep stage classification typically employs a single population-agnostic model, disregarding established demographic variations in sleep architecture. Sleep patterns, however, differ substantially across gender, age, and obstructive sleep apnea (OSA) severity, indicating that a onesize-fits all approach may be suboptimal for diverse clinical populations. In this paper, we propose a two stage training strategy based on demographic stratification and transfer learning framework. We first pretrains a convolutional recurrent model on the full population and then fine tunes it independently for demographic subgroups defined by gender, age, and Apnea-Hypopnea Index (AHI) severity according to the AASM clinical standard. Using the DREAMT dataset comprising 100 clinical subjects and 7 PSG channels, we evaluate 37 fine-tuned configurations across single-axis and two-way demographic combinations. Results demonstrate that 35 of the 37 fine-tuned models outperform the baseline, with Cohen's kappa improvements ranging from 0.9 to 12.9%. These findings indicate that stratified fine tuning tailored to specific patient demographics yields substantially more accurate sleep staging than a single generalized model, offering a practical and clinically grounded paradigm for personalized sleep assessment.

Recursive language-model loops often settle into recognizable attractor-like patterns. The practical question is how much injected text is needed to move a settled loop somewhere else, and whether that move lasts. We study this in 30-step recursive loops by separating the model from the context-update rule: append, replace, and dialog updates expose different histories to the same generator. The main result is that persistent redirection in append-mode recursive loops is memory-policy-conditioned. Under a 12,000-character tail clip, destination-coherent persistence plateaus near 16 percent and retained source-basin escape near 36 percent at dose 400; neither crosses 50 percent. Under a full-history protocol, retained source-basin escape crosses 50 percent near 400 tokens and saturates at 75-80 percent by 1,500 tokens; destination-coherent persistence first reaches 0.50 near 1,500 tokens (Wilson 95 percent CI [0.41, 0.61]). A four-step falsification battery (heterogeneity control, granularity sweep with hierarchical macro-merge, transition-entropy diagnostic, and long-horizon trajectory continuation) recasts the high-dose destination-coherent dip as a finite-horizon, endpoint-definition-sensitive feature rather than a stable structural asymmetry. Half the canonical magnitude is endpoint timing; the residual drops 73 percent from -0.143 at step 29 to -0.039 at step 79 under the frozen canonical cluster basis, bootstrap interval straddling zero. Replace-mode raw switching is near-saturated under the default protocol but largely reflects state-reset overwrite: insert-mode probes drop it to 12-32 percent. We report 37 experiments on gpt-4o-mini with within-vendor replication on gpt-4.1-nano. Recursive-loop evaluations should distinguish transient movement from durable escape, subtract stochastic floors, and treat context-update rules as safety-relevant design choices.

Gliomas are aggressive brain tumors that infiltrate surrounding tissue beyond the visible tumor margins observed on Magnetic Resonance Imaging (MRI). Predicting the spatial extent of this infiltration is essential for surgical planning and radiation therapy, yet existing deep learning approaches focus on segmenting the visible tumor rather than estimating infiltration risk in the surrounding tissue. This paper presents InfiltrNet, a novel dual-branch architecture that combines a convolutional neural network (CNN) encoder with a Swin Transformer encoder through cross-attention fusion modules to predict three-zone infiltration risk maps from multimodal MRI. A label generation strategy based on distance transforms is proposed to derive reproducible infiltration risk zones from standard Brain Tumor Segmentation (BraTS) annotations. InfiltrNet is trained with a combined Dice-CrossEntropy and boundary-aware loss augmented by auxiliary supervision heads at intermediate decoder levels. Extensive experiments on BraTS 2020 and BraTS 2025 demonstrate that InfiltrNet outperforms five established baselines. Explainability analysis using GradCAM++ and Occlusion sensitivity confirms that the model attends to clinically relevant peritumoral regions.

Evaluating a new language model means running it on a large, growing collection of benchmarks, at considerable cost. Much of this cost is redundant: arranged as a matrix of models by benchmarks, scores are strongly correlated and concentrate in a few dimensions, so a few well-chosen benchmarks largely determine the rest. This allows us to run a new model on a few benchmarks and predict the rest with a simple Gaussian model that copes with missing scores. The key question is which few to run. An answer comes from sensor placement, which chooses the observations that carry the most information, measured by entropy or mutual information. We argue that the objective should instead reward a selection for how well it predicts every benchmark. This leads to two coverage objectives: explained variance, the predictions' accuracy if the estimated correlations were exact, and facility location, which credits each benchmark with its best proxy among the selected ones. On eleven score matrices built from public evaluation results, the coverage objectives predict the scores of held-out models best wherever the correlations can be estimated. The information objectives fall behind because they favor benchmarks that the selection predicts poorly (entropy) or that the others predict well (mutual information), not benchmarks that predict the others. By contrast, facility location predicts practically as well as explained variance, holds up on the addition of newer models and under real costs, has classical guarantees (even under a budget), and gives readable selections.

Pneumonia remains a leading global cause of morbidity and mortality, particularly in low-resource settings where access to imaging, laboratory testing, and specialist care is limited. Clinical assessment relies on heterogeneous evidence, including symptoms, respiratory patterns, spoken descriptions, and chest imaging, making frontline screening inherently multimodal. However, many existing computational approaches remain unimodal and focus primarily on radiographs. In this work, we present MultiSense-Pneumo, a multimodal research prototype for pneumonia-oriented screening and triage support that integrates structured symptom descriptors, cough audio, spoken language, and chest radiographs. The system combines deterministic symptom triage, LightGBM-based acoustic classification, domain-adversarial radiograph analysis using ResNet-18, transformer-based speech recognition, and an interpretable late-fusion operator. Each modality is transformed into a normalized concern signal and aggregated into a unified screening estimate. The fusion weights are hand-specified and are treated as heuristic, interpretable parameters rather than learned or clinically optimized values. MultiSense-Pneumo is implemented with offline execution in mind on standard laptop-class hardware, but it is not presented as a deployment-validated or clinically validated diagnostic system. Experimental results demonstrate strong component-level performance of the radiograph pathway under synthetic domain shifts, while also highlighting important limitations, especially reduced abnormal-class recall for cough acoustics and the absence of paired end-to-end multimodal patient evaluation. MultiSense-Pneumo is therefore intended as a framework and component-level prototype for screening and triage research.

Machine unlearning aims to remove the influence of training data without retraining a model from scratch. Evaluation of vision-language models relies on forget accuracy (FA), retain accuracy (RA), membership inference, and similarity to a retain-only reference, but evaluation pipelines are rarely tested. We propose a validity-gated unlearning audit (VGUA) with three checks: the forget set must be memorized without duplicate leakage into the retain set, the scorer must measure correctness, and the training code must implement the intended procedure. Applying these checks reveals that random MMBench splits leak duplicate questions, a lenient scorer hides memorization, and two coding errors remove the signal while training completes. After correcting these problems, we evaluate 234 unlearned models using six methods, three learning rates, and three backbones. Each model is compared with a retain-only reference trained using low-rank adaptation (LoRA). The results depend on update strength. Among models with RA at or above 0.70, lower FA generally indicates closer agreement with the retrained FA-RA operating point. Aggressive updates overshoot this point, collapse RA, and reverse method rankings. Relative to the original model, 78% of healthy models move closer to the retrained model's scores, whereas 94% move farther from it in activation space. Membership-inference advantage does not reliably track retraining alignment. Under option-shuffled training, unlearned models answer 25% of reordered forgotten ScienceQA questions, compared with 6% for the retain-only reference. These results show that multimodal unlearning conclusions depend on pipeline validity, update strength, and the comparison criterion.