Papers

As Mixture-of-Experts (MoE) models scale toward hundreds of experts and higher top-$k$ routing, memory efficiency in distributed training becomes a critical bottleneck. Peak memory is dominated by the MoE block, not attention: every intermediate buffer in the MoE dispatch pipeline is individually scaled by top-k routing. The standard all-to-all dispatcher sends all routed tokens in a single collective step, requiring the full top-$k$-expanded buffer to be constructed at once. In this work, we propose RelayMoE, a ring-based MoE execution model that computes locally as expert weights or tokens circulate, avoiding full top-$k$-expanded dispatch buffers. RelayMoE selects between expert and token routing according to communication volume and overlaps transfers with computation. The ring structure naturally supports memory-efficient MoE recomputation during backward: each hop reconstructs expert intermediates, uses them to compute gradients, and releases them before the next hop. The saved memory supports longer sequences and larger batches, or retains more attention activations to reduce attention recomputation and improve training throughput. We evaluate RelayMoE on 30B$-$57B production MoE models and varied expert configurations. In single-layer MoE experiments, RelayMoE achieves a $2\times$ average speedup over Megatron-LM. In full-model training under the same GPU memory budget, it improves throughput by up to $2.02\times$ and extends the largest tested trainable sequence length by up to $2.85\times$.

Long-horizon LLM agents are frequently implemented using sparse mixture-of-experts (MoE) models, yet the co-design of agentic behavior and MoE structures remains underexplored. In this work, we comprehensively study the connections between agentic post-training and MoE expert selection. In off-the-shelf MoE models, we observe expert selection exhibits a specialized structure that naturally aligns with agentic trajectories. Specifically, expert routing overlaps more between turns where the agent performs semantically similar operations (e.g., READ, UPDATE) than between turns with differing operations. However, standard RL algorithms ignore this specialization, allowing the MoE routing to go uncontrolled during training, which empirically limit task performance and inference efficiency. To address this, we introduce a hierarchical routing control framework for agentic tasks. We explicitly encourage turn-level expert selections to align with agentic operations while regularizing token-level expert selections to maintain local consistency. To resolve stability issues that arise during post-training with the proposed methods, we further introduce an entropy-gated control mechanism. Overall, our routing control framework achieves over 10-point improvements in success rate on all evaluated benchmarks. These results demonstrate that agentic trajectory structure provides an effective signal for optimizing MoE capacity during RL post-training.

Modern science has moved decisively toward larger and more collaborative research teams. Yet an influential 2019 study reported that small teams are more disruptive than large teams, a conclusion difficult to reconcile with both the well-known trend toward increasing collaboration and the disproportionately large teams behind widely recognized disruptive research, such as work that has been awarded Nobel Prizes. Here we show that the reported disruptive advantage of small teams is largely an artifact of measurement rather than a fact about teams. Analyzing about 60 million scientific publications, we find that larger teams cite more references and that the standard disruption index declines mechanically as reference lists lengthen. Equalizing reference-count distributions across team sizes reverses the negative gradient within the index's own framework. Evaluating each paper against one reference at a time yields a reference-robust measure that distinguishes recognized breakthroughs from ordinary papers more sharply and more consistently than the standard disruption index across six independently curated benchmarks, including the Nobel Prize. Under this validated measurement, scientific disruption increases with team size across four scientific domains and in every decade from the 1970s to the 2010s. For scientific publications, we conclude that larger teams are more, not less, disruptive.

Learning with data in the form of quantum states is of current interest and has led to a variety of problems that boil down to interaction with the available data via quantum measurement and classical post-processing of observed classical outcomes. In quantum measurement PAC learning, one is given a sequence of unknown, prepared quantum states and classical labels, along with a hypothesis class of candidate measurements. The task is to select a measurement from the hypothesis class that minimizes a fixed notion of error in prediction of the classical labels via measurement of a new state by the selected hypothesis. In this work, we consider the advantage of interacting with the given data in the measurement learning framework using learning rules given by measurements that cannot be implemented using local operations and classical communication (LOCC), as opposed to single-copy learning rules. We provide a construction showing that there exist learning scenarios wherein single-copy learning rules are asymptotically suboptimal compared to optimal ones. We then show that learning rules based on entangled measurements enjoy at most a polynomial sample complexity advantage over single-copy learning rules in the PAC learning setting (under a natural joint measurability covering assumption).

Agent tasks require sequences of interdependent decisions. Autoregressive models support more flexible decision interfaces than conventional classifiers but incur the latency of token-by-token generation. Recent shared-prefix methods reduce this cost by reusing encoded context and scoring multiple decisions in parallel, but do not model decision dependencies or verify execution. We propose SharedKV-BT, where each active node of a behavior tree (BT) exposes stage-local fields and candidates, and Shared-KV scores the candidates in parallel and passes the selected decision to a separate execution system. We tested SharedKV-BT on robot manipulation, mobile navigation, and computer-use tasks. Across three tasks, SharedKV-BT made typed decisions 2.36-4.15 times faster than prompt-matched autoregressive decoding. On the manipulation task, node-local Shared-KV improved joint decision accuracy from 75% to 94% and closed-loop success from 0% to 60%. Fixed-score policy replay showed that stage gating prevented out-of-order actions and external postconditions prevented premature completion.

Universal object localization in X-ray security inspection is critical for automated threat detection in safety-critical venues. However, unlike everyday RGB images that dominate web-scale visual data, X-ray scans exhibit distinct color patterns, ambiguous boundaries, and compositional structures caused by volumetric superposition. These gaps hinder the direct zero-shot transfer of dense perception foundation models trained on web-scale RGB data. Moreover, annotated X-ray data is scarce and requires expert labeling, limiting both the training of generalizable X-ray native models and the adaptation of RGB foundation models for X-ray data via fine-tuning. Given these challenges, the bright promise of highly generalizable perception models, enabled by data scaling laws in the RGB domain, remains largely out of reach for X-ray inspection. To this end, we introduce LAO-X, a self-supervised adaptation framework that Locates Any Object in X-ray scans using diverse synthesized image–annotation pairs with granularity-aware supervision. LAO-X first designs a saliency-guided X-ray object mining module to separate diverse object instances, which are then used for physics-guided synthesis in the absorbance domain. LAO-X further incorporates an occlusion-controlled curriculum strategy to fine-tune a Segment Anything Model 2 (SAM2) localizer, progressively adapting it to X-ray scans with increasing object counts and overlap levels. Experiments on six X-ray benchmarks show that LAO-X substantially improves category-agnostic localization, achieving 2% to 23% mAP gains over SAM2 and X-ray specific baselines in heavily cluttered scenarios, entirely without human-annotated labels.

Time series foundation models (TSFMs) are trained on large collections of time series datasets that span various morphologies and domains. This setting exposes models to series whose scales – typical magnitudes of their values – can differ substantially. Affine scaling methods such as Reversible Instance Normalization (ReVIN) scale model inputs and reverse the transform before computing the loss. We show that this inversion multiplies each series' gradient by $b^p$ relative to loss on scaled targets, where $b$ is the scaling denominator (e.g., standard deviation) and $p$ is the loss degree. We call this scale-contaminated training (ScaleCon), because the scale of each series consequently becomes an importance weight, causing high-scale series to dominate training. For any scale-equivariant scaler and residual loss that is homogeneous of degree $p$, including MSE, MAE, and Quantile Loss, we prove that computing loss on scaled targets makes every mini-batch gradient and, consequently, the full optimization trajectory invariant to arbitrary independent rescaling of the training series, yielding scale-invariant training (ScaleIn). Notably, existing TSFMs use both objectives, with neither consistent reporting nor a common convention on how to compute training loss. We isolate the convergence disparity induced by ScaleCon and its correction under ScaleIn in controlled studies on synthetic and real data. In pretraining across four TSFM architectures, ScaleIn lowers MASE in all 24 architecture-benchmark comparisons, with average reductions across TSFMs of 18.8% on GIFT-Eval and 21.9% on the M-competitions. The gains extend to supervised neural forecasting, where it lowers MASE in 16 of 20 matched settings. Most existing time series forecasting pipelines can adopt ScaleIn with a one-line code change.

Security evaluation of learning-based systems requires more than just testing the system against a fixed collection of attacks. It requires adaptive mechanisms that can efficiently discover sets of inputs that induce model failure. We introduce ATLAS (Adaptive Trust-Regions for Latent Adversarial Searches), which is a query-based framework that discovers adversarial input sets for black-box learning systems. ATLAS casts attack generation as an active learning level set estimation problem then combines calibrated approximations with a local-global sampling architecture to find regions of the input space that contain adversarial examples. Once discovered, ATLAS is designed to sample points within these adversarial regions to build adversarial sets that accurately represent the state of robustness of the target model. When applied on toy experiments, we find that ATLAS is able to recover more of the adversarial region under a limited query budget than does previous work. When applied to standard and adversarially trained MNIST, CIFAR, and ImageNet model targets, ATLAS produces better representative attacks than other query-based black-box attacks (NES, SignHunter, BayesOpt). ATLAS represents an automated red-teaming framework that can be used for both analyzing the robustness of learning-based systems under development and continuous auditing to see how the robustness of a system changes over time.

Neuromorphic hardware combines memory (synapses) and computation (neurons) into the same silicon substrate to avoid the von-Neumann bottleneck. This way, neuromorphic computing aims to make computational neuroscience simulations and AI processing faster and more energy-efficient. The common crossbar architecture integrates a synapse matrix (analog, mixed-signal or digital) with a neuron array into a neurosynaptic core, whereas multiple cores are interconnected via a dedicated spike routing network. One of such neuromorphic architecture is the BrainScaleS wafer-scale system which allows the realization of very flexible network architecture, supporting both dense and sparse connectivity by highly configurable neurosynaptic cores. Yet, when network models from computational neuroscience are mapped to the BrainScaleS wafer, synapse loss can occur which means that for some model synapses no hardware synapse is available due to the restricted connectivity of the hardware. In this work we analyze the synapse loss when mapping uniform random networks to BrainScaleS both theoretically and empirically. We first develop a methodology to estimate maximum-sized networks without synapse loss based on probability distributions and the hardware's synapse routing architecture. Next, we adopt the methodology to predict the synapse loss for denser or larger network models. Then, we compare the predictions with results from actual runs of the mapping software: The results show the capabilities and limitations of the BrainScaleS system itself and spot shortcomings of the current mapping algorithms. The developed methodology can be helpful in multiple ways: to find the optimal settings when mapping a given network model to BrainScaleS, to act as a benchmark for the mapping software, and as tool for design space exploration for new hardware.

Multi-robot missions in infrastructure-denied environments frequently rely on reliable communication links between spatially separated locations. We propose a fully decentralized framework for constructing and dynamically maintaining communication networks without centralized topology planning or global positioning infrastructure. Driven strictly by local interactions, robots reconfigure local network topologies and adjust their physical positions to minimize overall network length while adhering to communication constraints. Formal analysis shows that our local reconfiguration operations guarantee continuous network connectivity, strictly decrease network length with every branch transfer, and bound worst-case performance to the shortest starlike tree. Embodied simulations and physical multi-robot experiments confirm that our approach forms networks near the length of centrally computed Euclidean Steiner trees. Additionally, the system dynamically adapts to moving targets and optimizes deployment by utilizing only necessary connectors, preserving excess robots for auxiliary tasks. This work enables autonomous swarms to self-organize adaptive ad hoc communication infrastructure in communication-denied environments, which could support applications varying from subterranean exploration to planetary missions.

Compliant mechanisms provide compact and intrinsic structural compliance for regulating physical interactions between mechanisms and environments. However, different tasks demand distinct stiffness characteristics, often requiring task-specific optimization and redesign due to limited geometric design space and inherent coupling among multiple stiffness components. This paper presents a Lego-like stiffness configuration approach using stackable planar compliant modules. Three complementary module geometries are introduced, with their stiffness characteristics further regulated through beam width, plate thickness, and module orientation. A unified stiffness model is established for quantitative analysis of individual and composed modules. Further, a two-stage optimization method is presented to achieve desired stiffness profiles, combining a genetic algorithm for configuration and sequential quadratic programming for parameter refinement. Experimental verification shows deviations below 6.5% for simulated stiffness. A flexible wrist is further developed as a representative implementation, exhibiting distinct compliant and dynamic responses under different stiffness characteristics. An optimized modular composition realizes prescribed stiffness values and maintains compliant obstacle interaction during high-speed motion at 1 m/s, with a maximum tested angular compliance of approximately $15^\circ$. The proposed framework provides a systematic approach for constructing flexible interfaces with task-specific stiffness characteristics.

We study list coloring of common independent sets of two matroids. We construct a graphic matroid $M_1$ and a partition matroid $M_2$ with common chromatic number two and common list chromatic number three, showing that the two parameters need not be equal. This resolves a question raised by Király, later stated as a conjecture by Aharoni, Berger, Guo, and Kotlar. We also show that if two strongly base-orderable matroids are each $2$-colorable, then their intersection is $2$-list-colorable.

Logic programming is increasingly used as the symbolic component of neurosymbolic AI systems. We survey the main rule-based languages in this setting, namely Datalog, answer set, and probabilistic logic programs, along four axes: semantics, expressiveness, neural integration, and evaluation mechanism. We analyse over 50 recent systems and applications, comparing formalism usage across four research areas: databases and programming languages, machine learning, vision, and robotics. We provide a decision matrix mapping application scenarios to required features and close by outlining open problems.

Agentic memory allows LLM agents to reuse past experience, yet retrieved memories can also distort inference even when they are benign, correctly stored, and appropriately retrieved. We study this failure mode, which we call memory over-reliance. Across benchmarks and memory architectures, we find that memory is useful when past experience transfers to the current task, but can become misleading when only part of the evidence transfers. Failures are strongest under partial query-memory overlap, a pattern further confirmed by controlled experiments thatvary the amount of overlapping evidence. Motivated by this finding, we propose MEMTRIM, a plug-and-play framework that indexes memory evidence at write time and controls its reuse at read time. MEMTRIM removes repeated or conflicting evidence while preserving useful memory-specific information, requires no retraining, and applies to both embedding-based and structured memory systems.Experiments show that MEMTRIM reduces memory overreliance while preserving the benefits of useful memory across models and memory settings.

Security operations centres and national incident-response teams defending critical infrastructure collect abundant threat data yet struggle to turn it into actionable intelligence. A malware sandbox produces detailed behavioural evidence, but as a large, unranked report whose confidence is unstated. We present CG-CTI, an operational pipeline that converts live sandbox output (CAPEv2) into STIX 2.1, correlates it in a knowledge graph with other critical-infrastructure sensors, and attaches to every intelligence object an explicit confidence status derived from provenance, cross-source corroboration, and observation durability. This status gates automated action: only corroborated intelligence is eligible for automated enforcement, while lower-confidence objects are routed to analyst review or kept as context. A grounded language-model stage then narrates the confidence-qualified evidence, where each statement either cites a supporting object or is marked unsupported, so fabricated references are removed before analyst review. We implement CG-CTI within the CYBERGUARD project, whose consortium includes Romania's national cyber-security directorate, and evaluate it against the live sandbox on a labelled malware corpus, measuring conversion validity, indicator yield, technique coverage, corroboration, enforcement eligibility, latency, and summary grounding. CG-CTI turns fragmented sandbox output into corroborated, confidence-ranked, and auditable intelligence for critical-infrastructure defence.

Long-term-memory agents can retrieve relevant information that is inadmissible for the current request because it belongs to another principal, violates policy, or reflects an incompatible lifecycle state. Recall and final-answer accuracy do not reveal this: a route can appear safe by missing required evidence, while a correct answer may follow inadmissible prompt exposure. We introduce a retrieval-admissibility verification framework that assigns each memory-query pair one of three statuses (admissible, inadmissible, or unresolved), compares routes at matched required-evidence recall with bounds for unresolved cases, and tracks memory IDs through prompt exposure while linking exposure to target-level disclosure. We evaluate its stages on separate, non-pooled populations. A post-hoc top-20 reanalysis of frozen rankings from two public long-term-memory benchmarks, RHELM and MemOps, covers 3,767 queries. All released anchors lie within trusted query namespaces; with within-namespace scores unchanged, off-namespace filtering cannot lower their ranks. Top-20 anchor recall increases from 0.432 to 0.533, 80% recall feasibility from 0.237 to 0.311, and exact similarity evaluations decrease by 98.3%. In a frozen 72-case development diagnostic, a released-metadata reference preserves required evidence, whereas neither text-only verifier detects violations under the 1% required-anchor false-denial limit. Across 1,523 paired benchmark-native cases, namespace routing is associated with judged-accuracy gains of 0.053-0.068 across three readers; recall also changes, so this comparison is observational. In 16 controlled exposure scenarios, only one of four reader-specific 95% confidence intervals excludes zero for relevant-inadmissible literal disclosure (+0.156, 95% CI [0.031, 0.312]). Results motivate separate verification of candidate support, admissibility, prompt exposure, and answer disclosure.

Metric conditioning can change which equilibrium attracts learning dynamics without moving the equilibria of a differentiable game. We study how to use state-dependent symmetric positive-definite (SPD) metrics to select among known equilibria by keeping a designated equilibrium attracting while making a rival unstable. Building on classical results on matrix stability under positive-definite multiplication, we characterize when such destabilization is possible and show how restricting the metric to act independently on each player limits the stability changes that can be achieved at differential Nash equilibria. We then combine a stabilizing metric at the selected equilibrium with a destabilizing metric at the rival through smooth interpolation, yielding a single metric field that preserves every equilibrium of the game while assigning the desired local stability types. We also derive a step-size condition for implementing the method in discrete time. The approach is validated on a continuous-commitment stag hunt and an entropy-regularized iterated prisoner's dilemma, where it selects the desired equilibrium in both settings.

Scientific search systems can find papers that are relevant to a question, but they generally do not assess the quality of the evidence that those papers provide. We present Kurate, a system that uses large language models (LLMs) to assess the quality of published studies. Kurate uses both the paper and its related documents (e.g., the study's trial registration and protocol), and links each of its judgments to the passage of text on which that judgment is based. We applied Kurate to a corpus of 4,347 papers (3,913 of which report randomized trials) and scored each paper on 8 dimensions of study design and reporting: specifically, statistical power, causal identification, preregistration, selective reporting, measurement validity, analysis prespecification, reporting transparency, and conflict of interest and funding. Across the corpus, we found that papers most often exhibited issues with statistical power, selective reporting, and analysis prespecification, although average quality differed between clinical areas. When compared against expert annotations of 60 held-out clinical-trial documents, the information Kurate extracted matched the expert label in 221/242 protocol scorepoints and 294/370 results-publication scorepoints, with AC1 0.94 and 0.81, respectively. Using a well-reputed, high quality clinical trial as a worked example, we show how a single paper's overall grade breaks down into separate judgments, with each linked to specific evidence from the trial's registration, protocol, and published report. Together, these results show that large-scale quality assessment of this kind is feasible, and that it can be used to address meta-scientific research questions.

Sparse attention mechanisms have become increasingly important for transformer models processing long input sequences due to their lower computational and memory complexity compared to full self-attention. Longformer achieves this through a sliding-window attention mechanism that produces a structured banded sparse attention matrix. However, existing sparse transformer accelerators primarily target attention generation or unstructured sparsity, leaving sparse matrix–dense matrix multiplication (SpMM) for structured sparse attention largely unexplored. This paper presents a pipelined FPGA architecture for accelerating banded SpMM in Longformer. The proposed design exploits the predictable sparsity pattern of Longformer's attention matrix through a custom row-wise storage scheme with implicit indexing, eliminating the overhead of conventional sparse matrix formats while enabling regular memory accesses. The architecture employs parallel processing elements, pipelined adder trees, and a dual-path computation strategy to maximize throughput and hardware utilization. Implemented in Verilog and evaluated on an RFSoC platform using Vivado 2024.2, the accelerator sustains one complete dot-product result per clock cycle after an initial latency of 11 cycles while maintaining power consumption below 2.9 W. Operating at 100 MHz, the design achieves over 100 million dot-product outputs per second, demonstrating the effectiveness of directly exploiting structured sparsity for sparse transformer acceleration.

This paper introduces Bayesian Continuation Model Predictive Control (BC-MPC), an adaptive method that replans near-term antenna pointings while continuing a feasible baseline sweep, for finding and establishing contact with spacecraft immediately after launch when their true positions are most uncertain. After separation, launch and deployment errors can leave the spacecraft sufficiently far from its predicted position that operators must actively search for it during ground contacts to establish communications. Each missed detection provides information about which orbital hypotheses remain plausible, but searching one part of the uncertainty region can consume the time needed to reach another. BC-MPC updates a particle-based orbital belief after missed detections and replans antenna pointing guidance while retaining feasibility. Its continuation helps preserve future search coverage that greedy and short-horizon methods can sacrifice for immediate acquisition gains. We compare BC-MPC against the operationally prevalent along-track sweep and additional search algorithms under common search-time budgets, initial uncertainty, and antenna constraints. Across 96 simulated deployments with matched uncertainty, BC-MPC improves acquisition probability over the along-track sweep and reduces restricted mean acquisition time. We additionally evaluate the method using pre-launch, post-launch, and initial acquisition ephemerides for 83 spacecraft from the Transporter-16 rideshare launch. Using either pre-deployment or post-deployment two-line element sets (TLEs) for planning, BC-MPC improves modeled acquisition probability over the along-track sweep by approximately 12 percentage points under the commonly assumed uncertainty model.

In this paper, we present a model-based Galerkin framework for output-feedback control of nonlinear systems with known dynamics. With a suitable actuator augmentation, the finite-dimensional realization preserves the actuator dynamics and the constant input matrix. We retain finite-order nonclosure as an explicit additive residual. The LTI part and this residual describe the lifted nonlinear dynamics exactly. The decomposition does not require an invariant subspace of observables and also applies to non-control-affine systems. We use regional bounds on the closure residual, actuator-realization defect, and output-reconstruction error in the standard $\mathcal{H}_\infty$ design. The proposed controller is a linear dynamic system driven only by the measured tracking error and requires no online lifting. Retaining the state coordinates gives direct bounds on the original state. We use these bounds to derive an a priori containment condition. Under this condition, the nonlinear closed loop is forward complete, and the state and tracking error satisfy explicit transient bounds and are uniformly ultimately bounded. The cart-pendulum example illustrates the nonlinear closed-loop guarantees and compares the controller with full-state backstepping. Although the output-feedback controller uses only the measured tracking error, it achieves almost the same nominal tracking RMSE as backstepping, with a lower peak tracking error.

Cyber threat analysis increasingly depends on evidence distributed across vendor advisories, vulnerability databases, and threat intelligence sources. Turning these fragmented observations into timely decisions requires models to connect technical severity with evolving exploitation evidence and available defensive actions. We present POLAR, an LLM-powered framework for synthesizing real-world cyber evidence into threat-centric assessments for prioritization and mitigation. POLAR first disentangles overlapping incidents and grounds each threat in source-linked evidence. For prioritization, it infers severity metrics from cyber evidence and combines the resulting assessment with temporally ordered exploitation signals to estimate near-term exploitation likelihood. For mitigation, it links the synthesized threat data to authoritative remediation knowledge and organizes applicable actions according to threat urgency and operational constraints. We evaluate POLAR on real-world vulnerability evidence collected from public resources and compare it with multiple baselines. Across heterogeneous incidents and zero-day settings, POLAR improves threat ranking and mitigation retrieval while producing evidence-linked intermediate assessments that support analyst inspection. The results establish evidence synthesis as a practical foundation for LLM-based cyber decision support across related security tasks.

E-textile design involves complex decisions across materials, sensor and actuator structures, fabrication, garment integration, and data processing. It typically requires iterative prototyping and testing, which are time- and labour-intensive, while few practitioners possess cross-disciplinary expertise across all relevant domains. To identify current bottlenecks and explore how Generative AI (GenAI) might support the design process, we conducted two half-day co-design workshops, one in Shanghai and one in Winchester, with practitioners from materials science, electronics, garment design, human-computer interaction, and manufacturing. Twenty practitioners participated in the Shanghai workshop; ten of them had prior experience in e-textiles and form the contributing sample analysed here. A further ten practitioners participated in the Winchester workshop. Participants mapped their own design pipelines, annotated bottlenecks, and proposed where GenAI could provide support. Rather than presenting a ranked list of opportunities, we report a process map that indexes each proposed GenAI role to the pipeline stage at which practitioners located it, together with the conditions on which they stated its usefulness would depend. Across both sites, practitioners consistently identified domain-specific operational barriers, including data scarcity, the disconnect between prototyping and manufacturing, and trade-offs in material-hardware integration. They also emphasized that the primary barrier to GenAI-driven e-textile design is not general model capability, but the lack of standardized, machine-readable representations of e-textile designs. Based on these findings, we identify four classes of domain-tailored AI tools that could support future e-textile design processes.

Maintaining a stable thermal environment is critical for spacecraft carrying precision optics, as temperature gradients and thermal expansion can alter the geometric alignment of optical components, compromise wavefront sensing, and degrade image quality even when active thermal systems provide real-time control. In particular, the temperature of the optical payload must be controlled to sub-Kelvin tolerances about the desired setpoint, a requirement that is often difficult to achieve with conventional reactive control methods. To this end, this paper develops a novel nonlinear thermal model for the optical system of a spacecraft, accounting for conduction, radiation, and the changing radiative environment induced by the spacecraft's orbit. Using this model, we develop a model predictive control (MPC) methodology that enables closed-loop thermal regulation of the optical compartment under realistic orbital conditions. Since only partial, noise-corrupted temperature measurements are available, an Unscented Kalman Filter is employed to estimate the full thermal state, yielding a certainty-equivalence MPC controller obtained via recursive online optimization. The thermal model is validated against a high-fidelity thermal simulator, and the proposed approach is compared against conventional PID control, the prevailing method for on-orbit thermal regulation.

Missing covariates are frequently encountered in supervised learning problems, and classical methods for estimation using such data use carefully chosen imputation schemes for missing data, or likelihood approximations that lead to nonconvex $M$-estimation problems. These methods and their relatives are suitable for scenarios in which the covariate distribution is known, and more broadly, have enjoyed tremendous success in linear models. But even in basic nonlinear problems such as logistic regression in moderate dimensions, such methods can experience drastic failure modes when the covariate distribution is unknown. Motivated by the need for reliable alternatives, we consider the problem of parameter estimation in logistic regression with missing covariates. Crucially, we operate in the assumption-lean setting where the covariate distribution is unknown (but bounded). We design a stochastic approximation method that is based on $Z$-estimation with a novel monotone operator, and establish that our algorithm is computationally efficient and achieves provable signal recovery at parametric rates under the hypothesis that covariates are missing completely at random. Our theory sharply characterizes the $\ell_2^2$ risk of the estimator in terms of the missingness profile, accommodating heterogeneous observation probabilities. Importantly, it shows that our method always outperforms the de facto “complete-case” estimator that ignores observations with any missing data. Even in the setting with homogeneous missingness (in which each covariate is observed independently with probability $q$), our bounds exhibit intricate and nonstandard dependence on $q$ that can yield significant improvements over using only complete cases. We complement our upper bounds with new information-theoretic lower bounds that show that this intricate dependence on $q$ is fundamental in a minimax sense.

Recent advances in penalty-based methods for stochastic bilevel optimization (SBO) have eliminated the need for second-order derivative oracles. However, for stochastic nonconvex-strongly convex bilevel problems, existing first-order methods typically rely on nested loops and/or large batch sizes for attaining $O(\epsilon^{-6})$ or $O(\epsilon^{-4})$ sample complexity under standard bounded-variance assumption or mean-square smoothness assumption. Achieving these rates with a single-loop penalty method and a constant batch size remains challenging due to a large penalty value needed for an accurate approximation. To address this challenge, we develop a stochastic SIngle-loop COnstant-Batch first-order penalty method (SICO) that combines two complementary ingredients. First, it performs one stochastic-gradient update per-iteration for both the original lower-level and penalized problems, with a projection that controls the separation between their iterates. Second, it applies an exponential moving average to stabilize the upper-level gradient estimator. We show that this combination achieves $ O(\epsilon^{-6}) $ sample complexity using only $O(1)$ stochastic-gradient samples per iteration under unbiased, bounded-variance stochastic gradients. Under the additional mean-square smoothness assumption on the lower-level stochastic gradients, the same algorithm improves the complexity to $O(\epsilon^{-4})$ also with $O(1)$ batch size. To the best of our knowledge, this is the first work to match the best-known convergence rate for fully first-order SBO methods using a single loop and a constant batch size. This result addresses an open problem posed in the literature.

Manual repair of program failures is time-consuming and disruptive for software developers, particularly during the pre-submit phase where test failures occur within continuous integration systems. While Automated Program Repair has seen significant advancement through Large Language Models, existing state-of-the-art techniques primarily focus on post-submit workflows, operating offline without the low-latency requirements necessary to assist developers in real-time within their flow before they switch context. In this paper, we introduce FlowAgent, an AI agent deployed at Google to automatically repair test failures in the pre-submit outer-loop workflow inside continuous integration systems. Integrated into Google's internal developer tools, Critique and Cider,FlowAgent utilizes a ReAct-style generate-and-validate loop, as well as rigorous pre-execution and post-execution abstention filters to ensure high-quality suggestions under strict latency constraints. Based on our case studies, FlowAgent is highly effective. First, a manual evaluation conducted on 195 real-world test failures demonstrated 67.18% accuracy in suggesting correct fixes. Following its Google-wide deployment, FlowAgent suggested fixes on 295,508changes, of which developers previewed 65,069 and applied 28,554. Developer feedback from interviews indicate that the agent is useful in suggesting correct fixes, integration of autonomous repair agents into industrial software engineering workflows is received well, while interesting challenges and opportunities still remain.

Model optimizations help improve inference performance and accuracy of ML workflows. However, relying on a single model to perform inference across all data batches often fails to maximize accuracy and thus overall performance. In many cases, alternate models could perform better on specific subsets of data where a primary model underperforms. Our experiments with real ML workflows indeed show that switching models improves workflow accuracy by up to 23%. Yet, current systems lack the ability to adaptively switch between models based on performance, forcing users to manually test models in sequence. We present FlexiFlow, a dataflow system that dynamically switches between alternate models when the current model exhibits low accuracy. FlexiFlow learns to rank models using a novel multi-armed bandit approach that accounts for model runtimes, probability of passing user-defined assertions, and the computational structure of the ML workflow. We show that the standard Thompson sampling approach is insufficient for switching models in ML workflows. In contrast, our proposed approaches are effective and scales to complex real-world ML workflows. Experiments show that switching models at runtime while reusing intermediate results provides higher accuracy, but also 48% efficiency gain compared to sequential workflow runs.

We study an online trading problem where a trader, given a sequence of i.i.d. prices drawn from a known distribution $F$ on $[0,1]$, must make irrevocable buy, sell, or hold decisions subject to storage constraints. We investigate achievable algorithmic performance measured in terms of regret, the difference between the expected profit of the hindsight optimal policy that knows the entire price sequence and an online algorithm. We analyze finite atomic and continuous distributions characterized by their local behavior around the median which we capture using a parameter $\beta$. The parameter $\beta$ quantifies how the mass of prices accumulates around the distribution median. We identify a new driver of algorithmic performance, demonstrating that median gaps coupled with an initial inventory level of zero can force regret scaling of $\Omega(T^{(\beta + 1)/(2\beta+4)})$ — establishing a novel spectrum of fundamental limits on algorithmic performance. We then study STARS, short for Storage Trading by Averaging Repeatedly across multiple Scenarios, which simulates possible future price scenarios to approximate the value-to-go function and make buy/sell/hold decisions. We show that STARS obtain near-optimal algorithmic performance (upto poly-logarithmic factors) across a broad range of distributions. In particular, it achieves $O(\log T)$ regret for finite atomic prices, $\widetilde{O}(T^{\beta/(2\beta+2)})$ for continuous distributions without a median gap and $\widetilde{O}(T^{(\beta+1)/(2\beta+4)})$ for continuous distributions with a median gap for $\beta \geq 0$.

Analog hardware platforms offer the potential to reduce energy consumption over digital architectures, but in order to succeed, large-scale analog systems must also be able to operate with or recover from the variability of their components. Towards this goal, we derive and demonstrate the lock-in equilibrium propagation (LIEP) training method. LIEP provides local gradient information for each component in an oscillatory network without separate forward and backward sweeps, potentially allowing for in-situ learning capabilities on analog oscillatory hardware platforms. We demonstrate that LIEP can be used both for ab-initio training as well as recovering performance when pre-trained parameters are perturbed. We show that LIEP can be formulated as a three-factor update rule, and suggest that although the method is currently only validated on shallow networks, alternate architectures may allow it to extend to deep and large-scale networks addressing complex tasks.