Papers

Reasoning-capable large language models solve hard problems by emitting long chains of thought, paying heavily in latency, GPU time, and energy. Casual inspection of their traces reveals extensive reformulation, verification, and circular self-reflection, yet how much of this deliberation is actually necessary has never been measured at scale or explained from first principles. This paper closes both gaps. We formalise reasoning redundancy directly in terms of the reasoning model itself: the redundancy of a correct trace is the largest fraction of its trailing segmented steps that can be truncated while $\pi$, forced to terminate thinking and emit a final answer, still produces the correct answer. A large-scale quantification across four frontier reasoning models and two mathematical benchmarks shows that step-level redundancy is consistently high – between 61% and 93% across the 8 (model, benchmark) conditions we study, with the median critical prefix equal to a single segmented step in six of the eight conditions – that the finding is robust to the choice of judge family, and that although $\rho$ decreases with problem difficulty on MATH-500, all four models remain substantially redundant ($\rho \in [46\%, 85\%]$) even on the hardest Level-5 problems. We then prove that this redundancy is a structural consequence of length-agnostic outcome rewards, not a model-specific artefact: under any such reward, no finite expected stopping time is optimal. The result holds regardless of RL algorithm, base model, data distribution, or whether the policy is obtained via RL or distillation; over-thinking is therefore not a bug to be patched in individual models but a structural property of how current reasoning models are trained. Code: https://github.com/zhiyuanZhai20/how-much-thinking-is-enough

Aggregate metacognitive quality scores mask within-model variation across MMLU benchmark domains. We administered 1,500 MMLU items (250 per domain, under an a priori six-domain grouping) to 33 frontier LLMs from eight model families and computed Type-2 AUROC per model-domain cell using verbalized confidence (0-100). Total observations: 47,151. Every model with above-chance aggregate monitoring showed non-trivial domain-level variation. Applied/Professional knowledge was reliably the easiest benchmark domain to monitor (mean AUROC = .742, ranked top-2 in 21 of 33 models); Formal Reasoning and Natural Science were reliably the hardest (one of the two ranked bottom-2 in 27 of 33 models). The three middle domains were statistically indistinguishable (Kendall's W = .164). A subject-level coherence analysis (within-domain similarity ratio = 0.95) confirms the six-domain grouping is a pragmatic benchmark taxonomy, not a validated latent construct. Within-family profile-shape clustering is significant for Anthropic, Google-Gemini, and Qwen (permutation p < .0001) but not DeepSeek, Google-Gemma, or OpenAI. Gemma 4 31B showed a +.202 AUROC improvement over Gemma 3 27B. Three models classified Invalid on binary KEEP/WITHDRAW probes produced normal profiles under verbalized confidence, confirming probe-format specificity. Bootstrap 95% CIs on 198 cells have median width .199. Split-half aggregate stability r = .893; profile-level split-half is weaker (grand median r = .184). These results show stable benchmark-domain variation obscured by aggregate metrics, and support benchmark-stage domain screening as a step before deployment in specific application areas.

Chain-of-thought (CoT) reasoning and reasoning-tuned models such as DeepSeek-R1 are commonly assumed to reduce shallow heuristic biases by thinking carefully. We test this on position bias in multiple-choice QA and find a different story: within any reasoning-capable model, per-question position bias scales with the length of the reasoning trajectory. Across thirteen reasoning-mode configurations (two R1-distilled 7-8B models, two base models prompted with CoT, and DeepSeek-R1 at 671B) on MMLU, ARC-Challenge, and GPQA, twelve show a positive partial correlation between trajectory length and Position Bias Score (PBS) after controlling for accuracy, ranging from 0.11 to 0.41 (all p < 0.05). All twelve open-weight reasoning-mode configurations show monotonically increasing PBS across length quartiles. A truncation intervention provides causal evidence: continuations resumed from later points in the trajectory are increasingly likely to shift toward position-preferred options (16% to 32% for R1-Qwen-7B across absolute-position buckets). At 671B, aggregate PBS collapses to 0.019, but the length effect still manifests in the longest quartile (PBS = 0.071), suggesting that accuracy gates the expression of length-driven bias rather than eliminating the underlying mechanism. We additionally find that direct-answer position bias is a distinct phenomenon with a different footprint (strong in Llama-Instruct-direct, weak in Qwen-Instruct-direct, and uncorrelated with trajectory length): CoT reasoning replaces this baseline bias with length-accumulated bias. Our results argue that reasoning-capable models should not be treated as order-robust by default in MCQ evaluation pipelines, and offer a diagnostic toolkit (PBS, commitment change point, effective switching, truncation probes) for auditing position bias in reasoning models.

The reconstruction of ocean subsurface temperature (OST) using satellite remote sensing data holds significant scientific value for advancing the understanding of ocean dynamics and climate variability. However, the scarcity of subsurface observations, combined with the high degree of nonlinearity and spatiotemporal heterogeneity in subsurface processes, poses substantial challenges to the accuracy and generalization capability of traditional reconstruction methods. To address these limitations, this study proposes an adaptive framework that could capture both vertical structural dependencies and temporal variation patterns of OST via spatio-temporal clustering. By incorporating this framework with various deep learning models, e.g., dual-path convolutional neural networks (DP-CNN), Attention U-Net, and Vision Transformer (ViT), the OST field can be accurately reconstructed at a global scale only using surface observations, i.e., sea surface temperature (SST), sea surface salinity (SSS), sea surface height (SSH), and sea surface wind (SSW). Experimental results demonstrate that multiple deep learning methods using the proposed framework largely outperform their original counterparts, yielding improvements in RMSE ranging from 12.4% to 27.2%. This study provides a reliable solution for subsurface temperature reconstruction, offering important implications for meteorological modeling and climate change assessment.

Owing to the recent advancements in wearable devices for health care, the importance of BP estimation without cuffs increases. Cuff technologies are inappropriate for continuous BP measurement due to their inconvenient usage, invasive character, necessity of calibration, large size, and inability to perform long-term monitoring. Normally, the algorithm used for cuffless BP prediction employs machine learning models that operate according to the data-driven approach. However, although they show high numerical accuracy, ML models do not provide any interpretability, resulting in poor physiological validity and clinical applicability. We propose a combination of Windkessel and ML models that incorporates the physical aspects into the latter one. It is performed by reformulating Windkessel into a form that will allow employing ML models. The result is a system of ODEs which can be used in the neural network. Thus, the inclusion of physical constraints improves the data-driven approach by making models consistent with physics, understandable, and robust. For illustration, we apply the described technique using a publicly available MIMIC-II database that we obtain from the UCI Machine Learning Repository.

Source-free domain adaptation (SFDA) provides a practical solution to cross-subject EEG decoding by adapting source-pretrained models to unlabeled target domains without accessing source data. However, existing SFDA methods rely solely on the limited internal knowledge of source-pretrained models, leading to inferior cross-domain generalization and unreliable pseudo-labels. Although EEG Foundation Models (FMs) pretrained on large-scale data exhibit strong generalizability, their potential in SFDA remains largely unexplored. To this end, we propose FUSED, a Foundation-guided Source-free EEG Decoding framework that integrates a large-scale FM with a compact Specialist Model (SM) via dual-branch co-adaptation. Specifically, we introduce a Co-adaptation mechanism equipping both branches with linear and prototype views, enabling cross-branch pseudo-label generation. Additionally, we design a Consensus Filtering Mechanism that exploits the FM's inherent stability to identify high-quality samples, along with a Two-Stage Pseudo-Label Refinement scheme to suppress error accumulation through cross-branch arbitration. Finally, we calibrate the FM's decision boundaries via mutual information maximization with the SM, followed by knowledge distillation from FM to SM, forming a principled calibrate-then-distill pipeline. To our knowledge, FUSED is the first work to leverage EEG FMs within the SFDA framework for cross-subject EEG decoding. Extensive experiments across three EEG paradigms, including motor imagery, emotion recognition, and SSVEP, demonstrate consistent state-of-the-art performance, validating the effectiveness of foundation-guided synergy for robust and privacy-preserving EEG decoding.

Accurate and continuous estimation of cognitive workload is fundamental to creating adaptive human-machine systems. However, designing architectures that balance representational capacity with computational efficiency has been challenging for practical deployment. This paper introduces 1BT, a One-Block Transformer for compact and efficient EEG-based cognitive workload assessment. The model aggregates multi-channel temporal sequences via a minimal latent bottleneck, using a single cross-attention module followed by lightweight self-attention. A controlled study involving 11 participants performing three cognitively diverse tasks (abstract reasoning, numerical problem-solving, and an interactive video game) was conducted with continuous EEG recordings across two workload levels. Systematic architectural analysis identifies the most compact configuration that preserves high performance, while substantially lowering computational cost. The final model achieves high workload classification performance with under 0.5 million parameters and 0.02 GFLOPs, paving the way for a design direction for real-time cognitive workload monitoring in resource-constrained settings.

We study the deterministic global optimization of trained Gaussian process posterior mean functions over hyperrectangular domains. Although the posterior mean function has a compact closed-form representation, its global optimization is challenging because it remains nonlinear and nonconvex. Existing exact deterministic approaches become increasingly difficult to scale as the number of training data points grows, leading to approximation-based methods that improve tractability by optimizing a modified (inexact) objective. In this work, we propose PALM-Mean, a piecewise-analytic lower-bounding framework embedded in reduced-space spatial branch-and-bound. At each node, kernel terms that are locally important are replaced by a sign-aware piecewise-linear relaxation in an appropriate scalar distance variable, while the remaining terms are bounded analytically in closed form. We show this hybrid approach yields a valid lower bound for the posterior mean, while limiting the size of the branch-and-bound subproblems. We establish validity of the node lower bounds and $\varepsilon$-global convergence of the resulting algorithm. Computational results on synthetic benchmarks and real-world application problems show that PALM-Mean improves scalability relative to representative general-purpose deterministic global solvers, particularly as the number of training data points increases.

Downhole drilling telemetry presents a fundamental labeling asymmetry: surface sensor data are generated continuously at 1~Hz, while labeled downhole measurements are costly, intermittent, and scarce. Current machine learning approaches for downhole metric prediction universally adopt fully supervised training from scratch, which is poorly suited to this data regime. We present the first empirical evaluation of masked autoencoder (MAE) pretraining for downhole drilling metric prediction. Using two publicly available Utah FORGE geothermal wells comprising approximately 3.5 million timesteps of multivariate drilling telemetry, we conduct a systematic full-factorial design space search across 72 MAE configurations and compare them against supervised LSTM and GRU baselines on the task of predicting Total Mud Volume. Results show that the best MAE configuration reduces test mean absolute error by 19.8% relative to the supervised GRU baseline, while trailing the supervised LSTM baseline by 6.4%. Analysis of design dimensions reveals that latent space width is the dominant architectural choice (Pearson $r = -0.59$ with test MAE), while masking ratio has negligible effect, an unexpected finding attributed to high temporal redundancy in 1~Hz drilling data. These results establish MAE pretraining as a viable paradigm for drilling analytics and identify the conditions under which it is most beneficial.

We study the community detection problem in the non-uniform hypergraph stochastic block model (HSBM), where hyperedges of varying sizes coexist. This setting captures higher-order and multi-view interactions and raises a fundamental question: can multiple uniform hypergraph layers below the detection threshold be combined to enable weak recovery? We answer this question by establishing a Kesten–Stigum-type bound for weak recovery in a general class of non-uniform HSBMs with $r$ blocks, generated according to multiple symmetric probability tensors. In the case $r=2$, we show that weak recovery is possible whenever the sum of the signal-to-noise ratios across all uniform hypergraph layers exceeds one, thereby confirming the positive part of a conjecture in (Chodrow et al., 2023). Moreover, we provide a polynomial-time spectral algorithm that achieves this threshold via an optimally weighted non-backtracking operator. For the unweighted non-backtracking matrix, our spectral method attains a different algorithmic threshold, also conjectured in (Chodrow et al., 2023). Our approach develops a spectral theory for weighted non-backtracking operators on non-uniform hypergraphs, including a precise characterization of outlier eigenvalues and eigenvector overlaps. We introduce a novel Ihara–Bass formula tailored to weighted non-uniform hypergraphs, which yields an efficient low-dimensional representation and leads to a provable spectral reconstruction algorithm. Taken together, these results provide a principled and computationally efficient approach to clustering in non-uniform hypergraphs, and highlight the role of optimal weighting in aggregating heterogeneous higher-order interactions.

Information handling practices of LLM agents are broadly misaligned with the contextual privacy expectations of their users. Contextual Integrity (CI) provides a principled framework, defining privacy as the appropriate flow of information within context-relative norms. However, existing approaches either double inference cost via supervisor-assistant architectures, or fine-tune on narrow task-specific data. We propose extracting normative simulacra (structured representations of norms and information flows) from fiction novels and using them to fine-tune LLMs via supervised learning followed by GRPO reinforcement learning. Our composite reward function combines programmatic signals, including task clarity (subsuming schema validity, construct discrimination, and extraction confidence), structural completeness, internal consistency, and context identification, with an LLM judge that evaluates whether the model's privacy reasoning is grounded in the held-out normative universe of the source text. To mitigate overfitting, we introduce per-completion contrastive scoring: each completion is evaluated against both the correct normative universe and a randomly selected wrong one, teaching the model to condition on context rather than memorize source-specific norms. We evaluate on five CI-aligned benchmarks spanning distinct societal contexts and ablate the contributions of RL and normative grounding. Across seven models, SFT introduces a conservative prior toward restricting information flow, improving recognition of privacy-relevant situations but not the correctness of privacy judgments. GRPO with normative grounding achieves the highest score on a law compliance benchmark and strongest correlation with crowdsourced human privacy expectations, demonstrating that fiction-derived normative simulacra can teach contextual privacy reasoning that transfers to real-world domains.

While standard flow-matching models transport noise to data uniformly, incorporating an explicit generation order - specifically, establishing coarse, low-frequency structure before fine detail - has proven highly effective for synthesizing natural images. Two recent works offer distinct paradigms for this. K-Flow imposes a hard frequency constraint by reinterpreting a frequency scaling variable as flow time, running the trajectory inside a transformed amplitude space. Latent Forcing provides a soft ordering mechanism by coupling the pixel flow with an auxiliary semantic latent flow via asynchronous time schedules, leaving the pixel interpolation path itself untouched. Viewed from the angle of improving pixel generation, we observe that forcing - guiding generation with an earlier-maturing auxiliary stream - offers a highly compatible route to scale-ordered generation without rewriting the core flow coordinate. Building on this, we propose Frequency-Forcing, which realizes K-Flow's frequency ordering through Latent Forcing's soft mechanism: a standard pixel flow is guided by an auxiliary low-frequency stream that matures earlier in time. Unlike Latent Forcing, whose scratchpad relies on a heavy pretrained encoder (e.g., DINO), our frequency scratchpad is derived from the data itself via a lightweight learnable wavelet packet transform. We term this a self-forcing signal, which avoids external dependencies while learning a basis better adapted to data statistics than the fixed bases used in hard frequency flows. On ImageNet-256, Frequency-Forcing consistently improves FID over strong pixel- and latent-space baselines, and naturally composes with a semantic stream to yield further gains. This illustrates that forcing-based scale ordering is a versatile, path-preserving alternative to hard frequency flows.

Safety and security are essential for the admission and acceptance of automated and autonomous vehicles. Deep neural networks (DNNs) are widely used for perception and further components of the autonomous driving (AD) stack. However, they possess several limitations, including lack of generalization, efficiency, explainability, plausibility, and robustness. These insufficiencies can pose significant risks to autonomous driving systems. However, hazards, threats, and risks associated with DNN limitations in this domain have not been systematically studied so far. In this work, we propose a joint workflow for risk assessment combining the hazard analysis and risk assessment (HARA) following ISO 26262 and threat analysis and risk assessment (TARA) following the ISO/SAE 21434 to identify and analyze risks arising from inherent DNN limitations in AD perception.

The rise of IoT devices and the uptake of cloud computing have informed a new era of data-driven intelligence. Traditional centralized machine learning models that require a large volume of data to be stored in a single location have therefore become more susceptible to data breaches, privacy violations, and regulatory non-compliance. This report presents a thorough examination of the merging of Federated Learning (FL) and blockchain technology in a cloud-edge setting, demonstrating it as an effective solution to the stated concerns. We are proposing a detailed four-dimensional architectural categorization that meticulously assesses coordination frameworks, consensus algorithms, data storage practices, and trust models that are significant to these integrated systems. The manuscript presents a comprehensive comparative examination of two cutting-edge frameworks: the Multi-Objectives Reinforcement Federated Learning Blockchain (MORFLB), which is designed for intelligent transportation systems, and the Federated Blockchain-IoT Framework for Sustainable Healthcare Systems (FBCI-SHS), elucidating their distinctive contributions and inherent limitations. Lastly, we engage in a thorough evaluation of the literature that integrates a comparative perspective on current frameworks to discern the singular nature of this research within existing knowledge systems. The manuscript culminates in delineating the principal challenges and offering a strategic framework for prospective research trajectories, emphasizing the advancement of adaptive, resilient, and standardized BCFL systems across diverse application domains.

Self-play has recently emerged as a promising paradigm for post-training Large Language Models (LLMs). In self-play, the target LLM creates the task input (e.g., a question), which it then addresses itself by producing a task output (e.g., an answer). A reward model evaluates the output, and the rewards are used to train the LLM, typically via Reinforcement Learning (RL). A key benefit of self-play for post-training LLMs is its minimal supervision costs: self-play avoids the need for high-quality input-output pairs traditionally constructed by humans or expensive proprietary models. Existing work, however, explores self-play only for verifiable tasks, such as math and coding, for which objective ground truth is available and easily checkable. In this paper, we seek to extend self-play to more realistic open-ended tasks. We propose POP, a self-play framework that uses the same LLM to synthesize evaluation rubrics along with each input-output pair. The rubric is used to evaluate outputs and train the model. Crucially, we ground the framework on a content-rich pretraining corpus to (1) enable an exploitable generation-verification gap and reduce reward hacking, and (2) prevent mode collapse. On Qwen-2.5-7B, POP increases performance of both the pretrained base model and instruction-tuned model on multiple tasks ranging from long-form healthcare QA to creative writing and instruction following.

Causal discovery through experimentation and intervention is fundamental to robust problem solving. It requires not just updating beliefs within a fixed framework but revising the hypothesis space itself, a capacity current AI agents lack when evidence demands representations they have not previously constructed. We extend the blicket detector paradigm from developmental science to test this capacity in AI agents equipped with architectural scaffolding that targets hypothesis-space restructuring. Our compositional architecture has two discrete components: context graphs, which structure exploration as typed state machines, and dynamic behaviors, which monitor for evidence that the current hypothesis space is inadequate and expand it at runtime. Across 1,085 experimental trials, these components make orthogonal contributions: context graphs drive reasoning quality within the post-switch hypothesis space, accounting for 94% of the accuracy gain, while dynamic behaviors drive reasoning eligibility by detecting regime changes and preventing premature commitment to outdated hypotheses.

We consider Decision-Focused Federated Learning (DFFL), a predict-then-optimize setting in which multiple clients collaboratively train predictive models for downstream linear optimization problems without exchanging raw data. Besides the data heterogeneity typical of standard federated learning, clients may also have different objective functions and feasible regions. Building on the SPO+ surrogate loss, we derive heterogeneity bounds that separate objective shift, measured through cost-vector distances, from feasible-set shift, measured through support-function and shape-distance terms. We show that, for general compact feasible sets, small objective perturbations can still induce nonvanishing decision-focused loss discrepancies, while strongly convex feasible regions yield sharper stability-based bounds. We then lift these pointwise bounds to a local-versus-federated excess-risk comparison, showing that federation is beneficial when the statistical advantage of pooling exceeds a client-specific heterogeneity penalty. Computational experiments on polyhedral and strongly convex problems confirm that federation is substantially more robust under strongly convex feasible regions. Finally, we evaluate a simple validation-based interpolation between local and federated DFFL models. This interpolation mitigates the theoretical tradeoff and reduces aggregate regret and worst-client harm in both synthetic experiments and a PJM energy-pricing case study.

We study replicable algorithms for stochastic multi-armed bandits (MAB) and linear bandits with UCB (Upper Confidence Bound) based exploration. A bandit algorithm is $\rho$-replicable if two executions using shared internal randomness but independent reward realizations produce the same action sequence with probability at least $1-\rho$. Prior approaches to this problem are elimination-based and, in linear bandits with infinitely many actions, rely on discretization, leading to suboptimal dependence on the dimension $d$ and $\rho$. We develop optimistic alternatives for both settings. For stochastic multi-armed bandits, we propose RepUCB, a replicable batched UCB algorithm and show that it attains a regret $O\!\left(\frac{K^2\log^2 T}{\rho^2}\sum_{a:\Delta_a>0}\left(\Delta_a+\frac{\log(KT\log T)}{\Delta_a}\right)\right)$. For stochastic linear bandits, we first introduce RepRidge, a replicable ridge regression estimator that satisfies both a confidence guarantee and a $\rho$-replicability guarantee. Beyond its role in our bandit algorithm, this may also be of independent interest in other statistical estimation settings. We then use RepRidge to design RepLinUCB, a replicable optimistic algorithm for stochastic linear bandits, and show that its regret is bounded by $\widetilde{O}\!\big(\big(d+\frac{d^3}{\rho}\big)\sqrt{T}\big)$. This improves the best prior regret guarantee by a factor of $O(d/\rho)$, showing that our optimistic algorithm can substantially reduce the price of replicability. This is the first linear-bandit algorithm with an optimal dependence on $\rho$ for large number of arms. Finally, we extend our framework to stochastic generalized linear bandits by developing RepGLM, a replicable penalized GLM estimator, and RepGLMUCB, a replicable optimistic algorithm for this setting.

Large language models (LLMs) are increasingly used for conversational clinical decision support, yet they conflate next token prediction with probabilistic decision making. We argue that this conflation reflects an architectural limitation: such systems lack explicit posterior tracking, controllable abstention thresholds, and auditable reasoning chains. We introduce MoBayes, a Modular Bayesian dialogue framework that separates reasoning from language. The LLM acts only as a language interface, parsing patient conversation into structured observations, while a Bayesian module performs probabilistic inference over these observations to update posteriors, select follow-up questions via expected-information-gain and determine when to stop or defer through calibrated decision thresholds. This design enables explicit posterior tracking, controllable selective decision-making, and replaceable population-specific statistical backends without retraining the language model. Across empirical and LLM-generated knowledge bases, MoBayes outperforms standalone frontier LLM doctors, including matched model-family comparisons where inexpensive sensor models paired with MoBayes exceed larger autonomous models at lower cost. The advantage persists under adversarial patient communication styles and across varying diagnostic scenarios. These results suggest that reliable conversational clinical decision support systems should separate probabilistic reasoning from language generation rather than scaling model size alone. Code is available at https://anonymous.4open.science/r/MoBayes/

As Large Language Models (LLMs) become increasingly popular, caching responses so that they can be reused by users with semantically similar queries has become a vital strategy for reducing inference costs and latency. Existing caching frameworks have proposed to decide which query responses to cache by assuming a finite, known universe of discrete queries and learning their serving costs and arrival probabilities. As LLMs' pool of users and queries expands, however, such an assumption becomes increasingly untenable: real-world LLM queries reside in an infinite, continuous embedding space. In this paper, we establish the first rigorous theoretical framework for semantic LLM response caching in continuous query space under uncertainty. To bridge the gap between discrete optimization and continuous representation spaces, we introduce dynamic $\epsilon$-net discretization coupled with Kernel Ridge Regression. This design enables the system to formally quantify estimation uncertainty and generalize partial feedback on LLM query costs across continuous semantic query neighborhoods. We develop both offline learning and online adaptive algorithms optimized to reduce switching costs incurred by changing the cached responses. We prove that our online algorithm achieves a sublinear regret bound against an optimal oracle, which reduces to existing bounds for discrete query models. Extensive empirical evaluations demonstrate that our framework approximates the continuous optimal cache well while also reducing computational and switching overhead compared to existing methods.

Rational design of covalent inhibitors requires simultaneously optimizing multiple properties, such as binding affinity, target selectivity, or electrophilic reactivity. This presents a multi-objective problem not easily addressed by screening alone. Here we present a machine learning pipeline for generating covalent inhibitor candidates using multi-objective reinforcement learning (RL), applied to two targets: epidermal growth factor receptor (EGFR) and acetylcholinesterase (ACHE). A SMILES-based pretrained LSTM serves as the generative model, optimized via policy gradient RL with Pareto crowding distance to balance competing scoring functions including synthetic accessibility, predicted covalent activity, residue affinity, and an approximated docking score. The pipeline rediscovers known covalent inhibitors at rates of up to 0.50% (EGFR) and 0.74% (ACHE) in 10,000-structure runs, with candidate structures achieving warhead-to-residue distances as short as 5.5 angstrom (EGFR) and 3.2 angstrom (ACHE) after further docking-based screening. More notably, the pipeline spontaneously generates structures bearing warhead motifs absent from the training data - including allenes, 3-oxo-$\beta$-sultams, and $\alpha$-methylene-$\beta$-lactones - all of which have independent literature support as covalent warheads. These results suggest that RL-guided generation can explore covalent chemical space beyond its training distribution, and may be useful as a tool for medicinal chemists working on covalent drug discovery.

The integration of single-cell proteomic data is often hindered by the fragmented nature of targeted antibody panels. To address this limitation, we introduce scpFormer, a transformer-based foundation model designed for single-cell proteomics. Pre-trained on over 390 million cells, scpFormer replaces standard index-based tokenization with a continuous, sequence-anchored approach. By combining Evolutionary Scale Modeling (ESM) with value-aware expression embeddings, it dynamically maps variable panels into a shared semantic space without artificial discretization. We demonstrate that scpFormer generates global cell representations that perform competitively in large-scale batch integration and unsupervised clustering. Moreover, its open-vocabulary architecture facilitates in silico panel expansion, assisting in the reconstruction of biological manifolds in sparse clinical datasets. Finally, this learned protein co-expression logic is transferable to bulk-omics tasks, supporting applications like cancer drug response prediction. scpFormer provides a versatile, panel-agnostic framework to facilitate scalable biomarker discovery and precision oncology.

Complex-Valued Neural Networks (CVNNs) have significant advantages in handling tasks that involve complex numbers. However, existing CVNNs are unable to quantify predictive uncertainty. We propose, for the first time, dropout-based Bayesian Complex-Valued Neural Networks (BayesCVNNs) to enable uncertainty quantification for complex-valued applications, exhibiting broad applicability and efficiency for hardware implementation due to modularity. Furthermore, as the dual-part nature of complex values significantly broadens the design space and enables novel configurations based on layer-mixing and part-mixing, we introduce an automated search approach to effectively identify optimal configurations for both real and imaginary components. To facilitate deployment, we present a framework that generates customized FPGA-based accelerators for BayesCVNNs, leveraging a set of optimized building blocks. Experiments demonstrate the best configuration can be effectively found via the automated search, attaining higher performance with lower hardware costs compared with manually crafted models. The optimized accelerators achieve approximately 4.5x and 13x speedups on different models with less than 10% power consumption compared to GPU implementations, and outperform existing work in both algorithm and hardware aspects. Our code is publicly available at: https://github.com/zehuanzhang/BayesCVNN.git.

Neural fields, also known as implicit neural representations (INRs), offer a powerful framework for modeling continuous geometry, but their effectiveness in high-dimensional scientific settings is limited by slow convergence and scaling challenges. In this study, we extend INR models to handle spatiotemporal and multivariate signals and show how INR features can be transferred across scientific signals to enable efficient and scalable representation across time and ensemble runs in an amortized fashion. Across controlled transformation regimes (e.g., geometric transformations and localized perturbations of synthetic fields) and high-fidelity scientific domains-including turbulent flows, fluid-material impact dynamics, and astrophysical systems-we show that transferable features improve not only signal fidelity but also the accuracy of derived geometric and physical quantities, including density gradients and vorticity. In particular, transferable features reduce iterations to reach target reconstruction quality by up to an order of magnitude, increase early-stage reconstruction quality by multiple dB (with gains exceeding 10 dB in some cases), and consistently improve gradient-based physical accuracy.

Large language models can be uncertain yet correct, or confident yet wrong, raising the question of whether their output-level uncertainty and their actual correctness are driven by the same internal mechanisms or by distinct feature populations. We introduce a 2x2 framework that partitions model predictions along correctness and confidence axes, and uses sparse autoencoders to identify features associated with each dimension independently. Applying this to Llama-3.1-8B and Gemma-2-9B, we identify three feature populations that play fundamentally different functional roles. Pure uncertainty features are functionally essential: suppressing them severely degrades accuracy. Pure incorrectness features are functionally inert: despite showing statistically significant activation differences between correct and incorrect predictions, the majority produce near-zero change in accuracy when suppressed. Confounded features that encode both signals are detrimental to output quality, and targeted suppression of them yields a 1.1% accuracy improvement and a 75% entropy reduction, with effects transferring across the ARC-Challenge and RACE benchmarks. The feature categories are also informationally distinct: the activations of just 3 confounded features from a single mid-network layer predict model correctness (AUROC ~0.79), enabling selective abstention that raises accuracy from 62% to 81% at 53% coverage. The results demonstrate that uncertainty and correctness are distinct internal phenomena, with implications for interpretability and targeted inference-time intervention.

Vision-language models (VLMs) are increasingly used in settings where sensitivity to low-level image degradations matters, including content moderation, image restoration, and quality monitoring. Yet their ability to recognize distortion type and severity remains poorly understood. We present DistortBench, a diagnostic benchmark for no-reference distortion perception in VLMs. DistortBench contains 13,500 four-choice questions covering 27 distortion types, six perceptual categories, and five severity levels: 25 distortions inherit KADID-10k calibrations, while two added rotation distortions use monotonic angle-based levels. We evaluate 18 VLMs, including 17 open-weight models from five families and one proprietary model. Despite strong performance on high-level vision-language tasks, the best model reaches only 61.9% accuracy, just below the human majority-vote baseline of 65.7% (average individual: 60.2%), indicating that low-level perceptual understanding remains a major weakness of current VLMs. Our analysis further reveals weak and non-monotonic scaling with model size, performance drops in most base–thinking pairs, and distinct severity-response patterns across model families. We hope DistortBench will serve as a useful benchmark for measuring and improving low-level visual perception in VLMs.

With the emergence of new evaluation metrics and attack methodologies for Membership Inference Attacks (MIA), it becomes essential to reevaluate previously accepted assumptions. In this paper, we revisit the longstanding debate regarding the correlation between MIA success rates and model generalization using an empirical approach. We focused on employing augmentation techniques and early stopping to enhance model generalization and examined their impact on MIA success rates. We found that utilizing advanced generalization techniques can significantly decrease attack performance, potentially by up to 100 times. Moreover, combining these methods not only improves model generalization but also reduces attack effectiveness by introducing randomness during training. Additionally, our study confirmed the direct impact of generalization on MIA performance through an analysis of over 1K models in a controlled environment.

Neural surrogates for stiff differential-algebraic equations (DAEs) face two barriers: soft-constraint methods leave algebraic residuals that stiffness amplifies into errors, and hard-constraint methods require trajectory data from stiff integrators. We introduce an extended Newton implicit layer that enforces algebraic constraints exactly and reduces fast dynamics to their quasi-steady-state values in a single differentiable solve. Embedded in a physics-informed DeepONet, the layer recovers all fast and algebraic states exactly from slow-state predictions, removes the per-window stiffness-amplification pathway, and yields a stiffness-scaled Implicit Function Theorem gradient absent from penalty methods. Cascaded implicit layers extend this to multi-component systems with provable convergence. On a grid-forming inverter (stiffness ratio of about 4712), extended Newton attains 1.42% error versus 39.3% (penalty) and 57.0% (standard Newton); augmented Lagrangian and feedback linearization diverged. Two independently trained models compose without retraining (0.72% to 1.16% error, exact constraint satisfaction). Cross-domain validation on the Robertson stiff DAE (stiffness ratio up to $10^5$) confirms generalization. Conformal prediction provides 90% coverage with automatic out-of-distribution detection.

Cement production is among the largest contributors to industrial air pollution, emitting ~3 Mt NOx/year. The industry-standard mitigation approach, selective non-catalytic reduction (SNCR), exhibits low NH3 utilization efficiency, resulting in operational inefficiencies and increased reagent costs. Here, we develop a data-driven framework for emission control using large-scale operational data from four cement plants worldwide. Benchmarking nine machine learning architectures, we observe that prediction error varies ~3-5x across plants due to variation in data richness. Incorporating short-term process history nearly triples NOx prediction accuracy, revealing that NOx formation carries substantial process memory, a timescale dependence that is absent in CO and CO2. Further, we develop models that forecast NOx overshoots as early as nine minutes, providing a buffer for operational adjustments. The developed framework controls NOx formation at the source, reducing NH3 consumption in downstream SNCR. Surrogate model projections estimate a ~34-64% reduction in NOx while preserving clinker quality, corresponding to a reduction of ~290 t NOx/year and ~58,000 USD/year in NH3 savings. This work establishes a generalizable framework for data-driven emission control, offering a pathway toward low-emission operation without structural modifications or additional hardware, with potential applicability to other hard-to-abate industries such as steel, glass, and lime.

We present MMCORE, a unified framework designed for multimodal image generation and editing. MMCORE leverages a pre-trained Vision-Language Model (VLM) to predict semantic visual embeddings via learnable query tokens, which subsequently serve as conditioning signals for a diffusion model. This streamlined design effectively transfers the rich understanding and reasoning capabilities of VLMs into the visual generation process. By obviating the need for deep fusion between autoregressive and diffusion models or training from scratch, MMCORE significantly reduces computational overhead while maintaining high-fidelity synthesis. MMCORE seamlessly integrates text-to-image synthesis with interleaved image generation, demonstrating robust multimodal comprehension in complex scenarios such as spatial reasoning and visual grounding. Comprehensive evaluations indicate that MMCORE consistently outperforms state-of-the-art baselines across a broad spectrum of text-to-image and single/multi-image editing benchmarks.