Papers

Runtime traces are becoming a central substrate for understanding agentic systems, yet interpretation has focused largely on what the agent did. Workplace agents face the complementary problem: interpreting the human activity that surrounds them. Hours of low-level events carry rich evidence about a user's state but are too granular to reason over directly, and flattening them into one stream or compressing them into a single embedding both treat "summarize the user's behavior" as if it had one correct answer. We argue instead that behavioral interpretation is resolution-dependent: the same trace should admit multiple addressable interpretations at different temporal resolutions. We construct a multi-resolution vocabulary of semantically normalized operators, recurring motifs, coherent episodes, and day-level rhythms, each preserving the structure salient at its own horizon. Applied to 667 million human-attributed events from a large commercial productivity suite (50,000 users, 100 organizations), it yields 120 operator types, thousands of motifs, 25 episode types, and five day-rhythm archetypes. We validate it on real telemetry: re-running the entire pipeline on a disjoint 2,000-user sample recovers the same taxonomy (structural stability), and on held-out users the full representation forecasts a user's next episode more accurately than a flat-operator baseline, a 17% relative macro-F1 gain (predictive validity), so the abstractions preserve future-relevant information rather than merely describe it. A controlled resolution ablation then shows that no single level is optimal across questions: different agent-facing questions about the same trace are best answered at different resolutions. Behavioral trace interpretation for agents should therefore be multi-resolution and query-conditioned: an agent should access the temporal grain a question needs, not one universal summary.

Vision foundation models (VFMs) are valuable in data-scarce domains such as surgery, where a single pretrained backbone can provide rich representations for many downstream tasks. Yet the dominant self-supervised pretraining paradigm uses only RGB images, leaving readily available complementary signals, such as depth maps, unused. This is a particular missed opportunity in surgery, where natural-image VFMs transfer poorly while the scene geometry is rich and informative. With strong off-the-shelf models now able to produce pseudo-labeled dense depth for any image corpus, we hypothesize that such signals can be folded into pretraining to learn better representations. We present DART, an RGB-D pretraining recipe that builds on DINOv2 with a simple modification: a pixel-space depth reconstruction objective applied to masked iBOT patches, supervised by pseudo-labeled depth. Depth is used only during pretraining, so fine-tuning and inference remain RGB-only. We find that this pixel-level reconstruction head improves representation quality rather than disrupting it. We further show that depth, which encodes scene geometry, is more effective as a target than alternative dense signals such as Canny edges, confirming that the gains stem from depth rather than added supervision alone. Across eight surgical benchmarks spanning segmentation, depth estimation, and image-level recognition, DART outperforms both natural-image and in-domain baselines, including a vanilla DINOv2 trained on identical data, improving dense prediction while also strengthening image-level understanding. More broadly, DART shows that freely available geometric pseudo-labels can strengthen foundation model pretraining without extra labels or added inference cost, pointing toward stronger backbones for surgery.

Unattended interactive autonomy - machines that step into danger in place of humans and complete tasks with human tools - remains a missing capability in mission-critical operations. These domains offer scarce training data and only onboard compute, yet deployed systems must face novelty without erasing prior competence. We introduce Continual Field-Adaptive Models (CFAMs), which learn efficiently in the lab and continue learning after deployment through autonomous, gradient-free, on-device updates. CFAM uses a complementary learning architecture with a frozen slow-learning component and a fast-learning Capsule Field. The slow component contains three cortices: Sensor, which maps multimodal input into 3D-grounded geometry; Reasoning, which decomposes tasks into skills and evaluates outcomes; and Action, which executes geometric skills. The Capsule Field stores field learning one-shot and gradient-free as Competence Capsules. Skill installation is few-shot in the lab and continual in the field; open-world novelty is outside scope. We evaluate CFAM across five embodiments: manipulator, quadruped, humanoid, quadrotor, and off-road vehicle. Baselines (pi0, CogACT, SpatialVLA) use the same in-house multi-embodiment dataset for physical-platform comparisons. CFAM reaches the operating point of a standard policy trained on the full prior-training dataset using 40% of the data, or 2.5x fewer trajectories. At test time, autonomous capture of verified near-OOD cases improves action success by 13.9 percentage points. In sequential simulation, backward transfer is -0.5 percentage points versus -11.4 for LoRA. CFAM therefore provides a bounded form of post-deployment physical intelligence: few-shot skill learning, autonomous field growth from verified near-OOD experience, and retention of prior competence.

Video-language benchmarks are usually constructed by the dataset authors without published reliability statistics, leaving the noise floor of the construct unknown. We argue that multimodal benchmarking benefits from methods taken from research communities that have already invested in strategies to ensure reliability. We illustrate the case with the Classroom Observation Protocol for Undergraduate STEM (COPUS): a 24-code multi-label observation instrument with a decade of peer-reviewed reliability literature. We recast COPUS as a video benchmark for multimodal foundation models, where it provides a dense set of structured labels (a 24-dimensional binary vector every 2 minutes across a 50-90 minute lecture), an externally validated vocabulary, and established literature that provides a per-code reliability target based on human evaluators. Annotations in our evaluation corpus are produced by a 5-person human-evaluator panel whose consensus matrix is our reference. We propose VISTA, a baseline that runs MiniCPM-V-4.5 over a dense sliding window, refines its per-window outputs with a lightweight multi-layer perceptron (MLP) head trained on top of the frozen backbone, and max-pools the resulting predictions onto the 2-minute COPUS grid. On three held-out chemistry lectures, VISTA reaches 80.1% restricted macro accuracy versus 74.9% for the zero-shot variant, with the largest residual errors on visually similar instructor codes and on rare audio-dependent codes. We characterize three systematic failure modes (audio-partial observability, fine-grained group-work discrimination, long-tail recall) and release the benchmark tooling, prompts and baseline code at https://github.com/ajfranck/VISTA.

The FitzHugh-Nagumo (FHN) system serves as a simplified model of neuronal voltage dynamics, capturing the activator-inhibitor structure behind both isolated action potentials and the rhythmic spiking seen across the brain. Exploring its 5D physiological parameter space is important for neuromodulation and mapping voltage recordings back to biophysics, yet classical finite-difference solvers make rapid parameter sweeps expensive. We train parameter-conditioned Fourier Neural Operators (FNOs) as fast, differentiable surrogates for the FHN voltage and recovery fields on a one-dimensional spatial domain, conditioning each Fourier layer on the parameter vector $\lambda = (D_u, D_v, a, b, \tau)$ via feature-wise linear modulation (FiLM). We apply a single bifurcation analysis that delimits the two distinct regimes the model spans, oscillatory (tonic firing) and excitable (action-potential propagation), and we train one operator in each. In the oscillatory regime the surrogate attains sub-$0.1\%$ relative $L^2$ error on both fields, runs nearly three orders of magnitude faster than the finite-difference baseline, generalizes uniformly across the parameter space, and extrapolates to low single-digit percentage errors outside of the training bounds. In the excitable regime the same operator accurately reproduces the firing threshold and the $c \propto \sqrt{D_u}$ conduction-velocity law and replicates full traveling pulses, fully capturing the excitable bifurcation structure rather than just smoothly interpolating fields.

As an increasing number of embedded platforms incorporate multiple processing units, shared resource contention induced unpredictable execution time poses a challenge for real-time system design. Memory bandwidth regulation is a popular mitigation approach, and MemGuard is the canonical example. Recently, MemPol introduced a new bandwidth regulation mechanism, which was compared with MemGuard. Specifically, they reported significant overheads for MemGuard, citing up to a 1.79x slowdown, to contextualize MemPol's comparative benefits. This report is meant to clarify and add the necessary nuance to the experiments carried out in that prior work. Specifically, we show that the MemGuard overheads presented in these prior evaluations were unintentionally amplified as the result of using a suboptimal configuration with an older version of MemGuard, wherein the benchmark under test was pinned directly to the master core responsible for handling global timer interrupts. By faithfully reproducing these specific experiments using a modern, decentralized implementation of MemGuard, we demonstrate that the actual execution overhead drops significantly under identical conditions. Consequently, when evaluated with a properly configured recent version, MemGuard exhibits an overhead that is highly comparable to MemPol's overhead. By revisiting these baseline metrics, this report provides an updated and comprehensive perspective required for future evaluations of memory bandwidth regulators.

Robots interacting with people must recognize not only explicit commands, but also social cues such as invitations, refusals, and unavailability. In real deployments, these cues must be inferred from noisy onboard perception under partial occlusion, changing viewpoints, and strict latency constraints. We present SocioGesture, a real-time adaptive social gesture perception system for human-robot interaction (HRI). SocioGesture uses a compact confidence-aware body-hand skeleton representation and a lightweight dual-stream model that fuses body motion with hand articulation for low-latency onboard recognition. To improve deployment robustness, we train the model with occlusion-aware skeleton corruption, exposing it to missing hands, occluded arms, and temporally unstable keypoints without increasing the inference cost. On a social gesture dataset collected in mixed indoor-outdoor HRI scenarios, SocioGesture achieves strong held-out-subject recognition, substantially improves robustness under structured joint occlusion, and runs in real time on a robot-mounted edge device. During deployment, uncertain interaction segments are saved for offline labeling and adaptation, enabling SocioGesture to expand its gesture vocabulary while preserving performance in the original classes. These results demonstrate a practical path toward robust, efficient, and adaptive social perception for interactive robots.

We present AgentiGrid, an agentic artificial intelligence (AI) framework that integrates large language models (LLMs) intelligence and high-performance computing (HPC) to streamline and accelerate the multi-scenario power flow studies. AgentiGrid is an autonomous decision-making agent that proposes parameter modifications, invokes analyses through HPC analysis toolkit ExaGO, interprets results, and determines subsequent actions. ExaGO provides multiple power flow applications that can perform deterministic, stochastic and security constrained optimal power flow analyses. AgentiGrid provides backends to multiple LLMs (OpenAI, Anthropic, Ollama, and Ollama cloud) augmented with context specific and task specific prompts. Key features include interactive mid-search steering, goal-type-aware post-search analysis, and concurrent variant exploration for power flow optimization. A Streamlit-based graphical launcher provides real-time visualization of iteration progress and generates reports in natural language. AgentiGrid is capable of autonomously converging transmission constrained alternating current optimal power flow in under 20 iterations, with near-perfect reliability

A key challenge in reliable LLM deployment is recognizing when uncertainty reflects irreducible variability in the task rather than limitations in the model's knowledge. In language tasks, a central source of such aleatoric uncertainty is input ambiguity or underspecification, where multiple interpretations remain plausible. Existing decomposition methods estimate aleatoric uncertainty by generating multiple clarifications of the input, querying the model for an answer under each clarification, and comparing the resulting answers. We argue that answers are not necessary for identifying ambiguity: they are often redundant, add avoidable cost, and can mislead through epistemic leakage. We support this claim theoretically, and propose a clarification-only approach that estimates this ambiguity-induced component directly from the space of plausible interpretations, without answers to the clarified inputs. Using ambiguity detection as an operational evaluation across three benchmarks, this direct approach improves AUROC (63.34 vs. 60.85), reduces computational cost by 4-26x in output tokens and 2.2-3.5x in API calls, and yields estimates with substantially lower correlation with epistemic uncertainty. Overall, our results suggest that ambiguity-induced aleatoric uncertainty is better estimated from the interpretation space than from the response space.

Human-artificial intelligence interaction can shape what scholarly inquiry becomes about. This study examines research-object formation in two isolated scholarly trajectories developed by the same expert researcher with Claude and ChatGPT under matched starting conditions. Both used the same five-session, 843-turn project corpus, frozen research problem, initial prompt, publication objective, and conduct rules. The subsequent scholarly-development conversations were reconstructed ex post through consequential-event analysis, a Socioduality relational-process overlay, and tracing into completed manuscripts. Both trajectories shifted the continuity problem from recall toward usability, then developed different research objects: continuity labour and distributed, evolving project state. Failed analytical units, rejected explanations, counterexamples, and changes in analytical scale shaped these objects and propagated into research questions, methods, findings, evidence logic, and theoretical contributions. Consequential scholarly change, reciprocal relational continuity, and local substantive re-formation had distinct boundaries; continued interaction did not itself establish epistemic endorsement. The findings demonstrate empirically constrained research-object formation under matched scholarly starts. For systems supporting sustained scholarly knowledge work, they motivate recoverable analytical provenance, inspectable proposal status, and source-faithful contribution records.

Digital materials fabricated by multi-material 3D printing are designed as controlled mixtures of stiff and compliant constituents, yielding effective responses that span more than an order of magnitude in apparent stiffness and exhibit strongly nonlinear, composition-dependent, and rate-dependent dissipative behavior. Classical finite-strain viscoelastic models represent such behavior with closed-form strain energy functions for equilibrium and non-equilibrium stresses as well as evolution of internal variables, which may limit flexibility when a single constitutive model is expected to generalize across materials and loading rates. Here, we present a data-driven multi-material constitutive modeling framework that generalizes a formulation by Bergström and Boyce. The proposed framework retains the structure of the classical model, namely multiplicative kinematics, invariant-based strain-energy functions, and a scalar dissipative evolution law directed along the normalized nonequilibrium deviatoric stress. For the equilibrium branch, the data-driven discovery framework either directly predicts closed-form model parameters as functions of composition or automatically constructs a polyconvex strain-energy function using neural ordinary differential equations (NODEs). The nonequilibrium branch kinetics are learned similarly, either by directly identifying closed-form parameters across compositions or by using appropriately constrained artificial neural networks. Using multi-rate uniaxial compression data across multiple material compositions, we show that the proposed formulation captures rate-dependent stiffness and hysteresis across compositions while preserving thermodynamic consistency.

We introduce Mitra-v2, a tabular foundation model that delivers state-of-the-art performance on real-world classification and regression problems, from credit-risk scoring and clinical prediction to equipment-failure detection and house-price estimation. Mitra-v2 is trained only on synthetic data, with a pretraining distribution that is much larger and more diverse than Mitra-v1's. Built on a small 2D Transformer backbone, Mitra-v2 supports longer contexts and larger feature spaces. Improved optimization lets it learn from this larger task distribution. We evaluate Mitra-v2 on the TabArena and TALENT benchmarks, comprising more than 300 real-world datasets under two evaluation protocols. On the full TabArena benchmark, Mitra-v2 delivers state-of-the-art performance at the level of the industry-scale TabFM and EXAONE Tabular models, while surpassing TabPFN-3 by a wide margin in both classification and regression. Mitra-v2 matches the 1.6B-parameter TabFM with only 5% of its size (77M parameters), delivering frontier performance at a fraction of the cost. On TALENT, Mitra-v2 remains among the leading models, clearly outperforming TabPFN-3 and TabICLv2. It also ranks first on classification tasks with more than ten classes, even though it was pretrained only on tasks with at most ten classes. These results make Mitra-v2 one of the strongest and most broadly applicable open tabular foundation models released to date. We release the model weights, the inference and fine-tuning code, and our evaluation results under the Apache-2.0 license.

A critical challenge in deploying Large Language Models (LLMs) is developing reliable mechanisms to estimate their confidence, enabling systems to determine when to trust model outputs versus seek human intervention. We present a Calibrated Reflection approach for enhancing confidence estimation in LLMs, a framework that combines structured reasoning with distance-aware calibration technique. Our approach introduces three key innovations: (1) a Maximum Confidence Selection (MCS) method that comprehensively evaluates confidence across all possible labels, (2) a reflection-based prompting mechanism that enhances reasoning reliability, and (3) a distance-aware calibration technique that accounts for ordinal relationships between labels. We evaluate our framework on diverse datasets, including HelpSteer2, Llama T-REx, and a proprietary conversational dataset, demonstrating its effectiveness across both conversational and fact-based classification tasks. This work contributes to the broader goal of developing reliable and well-calibrated confidence estimation methods for LLMs, enabling informed decisions about model trust and human judgement.

An approval committee is Hare-core stable if no coalition meeting the Hare quota can strictly improve by moving to another candidate set. Whether every approval election has such a committee remains open. We prove nonemptiness when each voter disapproves at most two candidates, with no bounds on the numbers of candidates, seats, or voter types. The result also permits arbitrary positive rational voter weights. Our deterministic rule represents a committee by its missing set. It first maximizes weighted coverage of two-candidate disapproval sets and then maximizes total disapproval incidence. An exact coverage inequality excludes targets one seat below the committee. The incidence objective excludes unanimous equal-size targets, while targets of size at most $k-2$ cannot improve any voter. Two implementation-level independent verifiers audit overlapping finite grids. The symbolic proof, not this bounded enumeration, establishes the theorem's unbounded quantifiers. The argument identifies complement-side coverage as a tractable mechanism for a broad parameter range within a sharply defined preference domain.

This paper considers a downlink communication framework comprising a simultaneously transmitting and reflecting reconfigurable intelligent surface (STAR-RIS)-aided by orthogonal time frequency space (OTFS) and non-orthogonal multiple access (NOMA) technologies. Further, delay-Doppler mobility in such frameworks renders classical alternating optimization impractical for per-coherence interval reconfiguration. To mitigate such issues, the STAR-RIS phase-shift and energy-splitting design is formulated as a constrained, non-convex sum-rate maximization problem with closed-form maximum ratio transmission beamforming and fixed NOMA power allocation. To circumvent the per-interval re-optimization burden, a deep reinforcement learning (DRL) approach is adopted that maps observed channel realizations to STAR-RIS configurations through a single forward pass. Specifically, Beta-Space Soft Actor-Critic (SAC-BSE), a maximum entropy DRL agent, is proposed. Simulation results, with two NOMA-multiplexed users on each STAR-RIS branch, confirm rapid convergence, limit the sum-rate degradation to roughly 10% across a 128-fold user-speed range, and yield consistent gains over OTFS-only, NOMA-only, STAR-RIS-only, fixed-split, and mode-switching baselines as transmit power and the number of STAR-RIS elements increase.

Static application security testing (SAST) tools help developers find vulnerabilities before deployment, but false positives create substantial triage effort. We study whether CodeQL false positives in Java security analysis form recurring, explainable patterns that can be reduced by refining the analysis. We run CodeQL's Java security query suite on 167 CVE instances from 110 projects, focusing on the ten queries with the highest false positive rates. We manually review 500 sampled false positive paths and locations and construct a source-level taxonomy. The five categories are Missed Path Constraint or Sanitization (36.6%), Benign Execution Context (29.4%), Missing Trust Boundary Modeling (27.6%), Imprecise Concurrency Modeling (5%), and Imprecise Sink Modeling (1.4%). Guided by these findings, we implement CodeQL refinements that detect and filter recurring false positive patterns at the query level. The refinements remove 81.8% of reviewed false positives. Across the full selected-query dataset, they remove 15.8% of reported paths and locations while retaining 7 of 8 true positives. This shows that many false positives can be reduced in the analysis, although fixed refinements often depend on project-specific context. To address this generalization gap, we evaluate whether agentic coding tools can adapt refinement patterns to new projects. Given our patterns as templates, the two tools succeed on 56% and 62% of tasks, with query compile-pass rates above 90%. Without this guidance, both succeed on only 28%, while compile rates fall to 30-36%. These results support a refinement-oriented SAST workflow in which recurring false positives are modeled in CodeQL queries and automatically adapted to different project contexts, reducing repeated triage.

I present a matrix-free symmetric interior penalty discontinuous Galerkin method for adaptively refined Cartesian meshes on GPUs. At non-matching interfaces, auxiliary shadow cells represent the adjacent coarse polynomial on the fine level. This approach converts non-matching interfaces into matching faces, permitting uniform face evaluation throughout the mesh. I prove that this construction is equivalent to the standard non-matching formulation. Furthermore, this representation yields a local geometric multigrid method in which frozen shadows provide inter-level boundary data and carry residual contributions to coarser levels. A primal–dual pairing eliminates shadow assembly from the Krylov iteration. Numerical experiments with cubic elements ($p=3$) on an NVIDIA A100 show stable multigrid convergence under increasing refinement depth and efficient GPU execution.

Prompt injection is widely recognized as a major security threat to AI agents that interact with untrusted external data, such as websites, documents, and emails. Prior work has shown that, in the text domain, black-box prompt injection can achieve near-perfect attack success rates (ASRs). In the image domain, however, existing visual prompt injection methods are substantially less effective in attacking frontier commercial VLMs for materially harmful behavior. Achieving such outputs is hard because it requires a long and/or format-compliant target string, such as a precise, parseable native tool call with exact function names and arguments. We present Repeat-After-Me, a black-box adaptive visual prompt injection attack that can reveal personally identifiable information or make malicious tool calls. Across both open-weight and commercial frontier VLMs, including Qwen3.6-27B and GPT-5.5, our method achieves ASRs exceeding 82% and 47%, respectively, under a realistic setting in which the benign user prompt is semantically unrelated to the injected task and does not verbally authorize it. Our optimized injection has non-trivial attack transferability across commercial VLMs and benign samples. We show our attack works in cases where adaptive textual prompt injection fails. In a real-world OpenClaw agent connected to Discord, an untrusted user can use a minimally injected image from our attack to overwrite TOOLS.md, enabling future sensitive behaviors like remote code execution and secret exfiltration. We discuss potential defenses.

Language generation is almost universally treated as a sequential process: autoregressive models emit one token at a time, while diffusion language models replace token-level seriality with a long trajectory of iterative refinement. In this work, we introduce PlaidQ, a 0.7B continuous diffusion language model for code generation, and show that its trajectory can be aggressively distilled into only a few denoising steps—or even one, enabling efficient code generation. PlaidQ repurposes a pretrained autoregressive model as a bidirectional denoiser over continuous token embeddings. We distill PlaidQ with distribution matching for few-step generation and paired-trajectory supervision for one-step generation. At matched model scale, PlaidQ is competitive with discrete diffusion language models on code generation. Distillation then shifts the quality–compute frontier: a 16-step student reaches 31.78 and 40.49 pass@10 on HumanEval and MBPP+, surpassing the same PlaidQ teacher sampled for 512 steps. At the extreme, paired-trajectory distillation achieves 7.07 pass@1 on HumanEval with a single denoising step, producing functionally correct programs. Together, these results establish continuous diffusion as a viable path to few-step and one-step code generation. Broadly, continuous diffusion is not merely another representation for language: it provides an interface through which language models can inherit the acceleration and distillation machinery of continuous diffusion modeling. Training and inference code and model checkpoints are available at https://github.com/pengzhangzhi/plaidq.

We introduce a reduced-order learning framework that couples a convolutional autoencoder to a structured latent neural ordinary differential equation. A learned scalar generates the latent dynamics through an antisymmetric operator and a positive-definite dissipative operator. The antisymmetric component permits motion along scalar level sets, while the dissipative component decreases the scalar. Adding a coercive quadratic term gives bounded continuous-time latent trajectories and approach to the scalar's critical set under the stated assumptions. The scalar provides a structural constraint rather than an approximation of physical energy. The encoder, decoder, scalar, and operators are trained jointly on short snapshot windows without time-derivative supervision, physical-energy labels, or dissipation penalties. At inference, one initial state is encoded and evolved continuously in latent space, with decoding only at requested times. We demonstrate the framework on micromagnetic magnetization trajectories from NIST $\mu$MAG Standard Problem 4 and compare it with dissipative-only dynamics and an unstructured latent neural differential equation. Repeated training runs show competitive observed-horizon accuracy, while selected longer rollouts illustrate improved retention of approximately stationary states relative to the unstructured baseline. These results motivate scalar-generated latent dynamics as an inductive bias for reduced-order prediction and other learning tasks where bounded evolution and equilibrium behavior are important.

Humans can learn and generalize novel concepts from sparse data because they express knowledge in rich structural formats. In this paper, we propose that programs are a strong candidate for universal representation of concepts. We review computational models of concept learning that use programs as their concept representation and evaluate their contribution toward a universal representational language.

Merging a LoRA adapter into its base model is standard deployment practice: it removes the runtime adapter's per-forward overhead and leaves a single standalone checkpoint any serving stack can load. On a native 4-bit microscaling checkpoint (NVFP4, MXFP4) that step stops being free. The merged weights must be written back through a quantizer, which re-derives the checkpoint's discrete E2M1 code plane (roughly 90% of the artifact's bytes), so the deployed artifact becomes coupled to one quantization convention, and every later code-touching event in its lifecycle can move it. Done naively the step is worse than fragile: it deletes the adaptation, by up to 39 pp, because against an already-on-grid base the reconstruction optimum is that base. Scale-QLoRA instead adapts only the native per-block scale field, trains those scales on the deployment grid, and freezes every E2M1 code. Within a fixed native format, scale grid, block layout and code plane, merging is then a bit-exact identity and the merged artifact is code-invariant. Across four models and four tasks, Scale-QLoRA and merge-aware QAT-LoRA are both accuracy-lossless, so we claim no accuracy ordering between them; they differ structurally, in that QAT-LoRA re-derives the code plane through a quantizer while Scale-QLoRA preserves it exactly. That difference is what the lifecycle prices: nearest-rounding implementations disagree by about a point on the measured task, and more extreme rule mismatches can drive the weight-space artifact to ~0%, which we report as a sensitivity bound rather than a deployment frequency. Preserving the code plane also drops the weight-space straight-through estimator from training (3.9x per step on the dense 8B model) and enables exact rollback, code-plane deduplication, and a ~125x faster scale-only task swap.

Existing Conditional Flow Matching (CFM) formulations describe transport progress using an explicit interpolation coordinate, commonly interpreted as time, assuming that a single global variable adequately represents a sample's position along the generative trajectory. In restoration tasks, however, transport progress is sample-dependent because the initial distribution may exhibit varying statistical dependencies with the target distribution. Thus, samples at the same interpolation coordinate can differ substantially in degradation level, distance to the target distribution, and restoration difficulty. We investigate whether signal representations learned by discriminatively trained models provide a meaningful description of generative transport state in CFM-based restoration. Through systematic latent-space analysis, we show that discriminative representations organize according to degradation severity and follow a consistent trajectory toward the clean-data manifold during generation. Motivated by these observations, we introduce the Discriminative Flow-State Hypothesis, which posits that discriminative representations encode a transport state governing generative restoration. Based on this hypothesis, we propose Discriminative Flow Matching, which conditions the Flow-Matching velocity field on Discriminative Flow-State Representations rather than explicit time coordinates. Experiments on speech enhancement and image denoising show that these representations characterize restoration progress, enable adaptive inference, and consistently outperform CFM and diffusion-related baselines. Our findings suggest that discriminative representations provide an effective state-aware alternative to explicit time conditioning and offer a novel perspective on the relationship between discriminative and CFM-based generative modeling.

Distributed quantum computing (DQC) offers a promising approach to scale quantum computing by overcoming the resource limitations of a single quantum processor. However, inter-node communication remains a major bottleneck of DQC due to inefficient and error-prone entanglement distribution. Optimizing inter-node communication can not only reduce the amount of entanglement resource needed to execute a quantum circuit but also improve execution speed and accuracy of the results. This paper proposes DPRQ, a qubit routing algorithm for minimizing inter-node communication in distributed quantum circuits divided into collective communication blocks. Unlike current approaches that utilize greedy block-level qubit routing strategies, DPRQ employs a dynamic programming-based technique focused on global circuit-level optimization, while capturing inter-block dependencies. We evaluated DPRQ on four sets of quantum circuits and a variety of DQC configurations. The results demonstrate that DPRQ's innovative routing strategy achieves an average of 24.40% reduction with a maximum of 85.06% reduction in inter-node communication, when compared to the state-of-the-art collective communication-based DQC compiler QuComm.

Designing and authoring high-performance custom kernels for accelerators is a complex task that requires deep hardware-level expertise. Large Language Models (LLM) can be leveraged together with real-time compiler feedback to build agentic systems for kernel generation. In this work, we present MaxKernel, a multi-agent system that implements three distinct paradigms for TPU kernel development: (1) a Human-in-the-Loop (HITL) agent for collaborative, step-by-step design; (2) an Autonomous (Auto) agent that executes a fully automated, metric/trace-driven optimization loop; and (3) a Graph-Based Autonomous Search that scales the Auto agent for global exploration of the design space. All three paradigms leverage a shared pool of specialized sub-agents to handle planning, implementation, self-debugging, testing, and hardware profiling. We evaluate MaxKernel on JaxBench, a comprehensive suite of 50 diverse kernel tasks for TPUs, alongside complex, real-world workloads from state-of-the-art open-source models. We demonstrate that MaxKernel consistently generates highly optimized implementations, matching expert hand-tuned baselines and delivering significant performance across the benchmark. Our agent is open-sourced and available https://github.com/AI-Hypercomputer/accelerator-agents/tree/main/MaxKernel.

Semantic search is widely deployed in modern AI systems, but protecting both data contents and access patterns remains challenging. The current state-of-the-art system, Compass, achieves oblivious semantic search by building an optimized ORAM over HNSW graphs. However, even with aggressive optimizations, it still incurs large overheads. Closing this performance gap is fundamentally difficult: Compass has already removed most cryptographic overheads, leaving ORAM accesses as the dominant cost, which are constrained by well-known Omega(log N) bandwidth lower bounds. Our key insight is that traditional ORAM overhead stems from the assumption of limited private memory, whereas modern GPU TEEs provide large private memory (Pmem) that blinds internal access patterns (Hunt et al., NSDI '23). This shift opens a new design space. We therefore propose Hoss, a first-of-its-kind oblivious semantic search system with a heterogeneous CPU-GPU TEE architecture that supports fast, scalable search with low cost of ownership. In Hoss, the GPU TEE's large Pmem hosts the hot-path HNSW traversal, while the lower layers of the graph, if they exceed GPU capacity, are offloaded to CPU TEEs. The system invokes oblivious primitives only when accessing these lower layers. The availability of large Pmem also enables new optimization opportunities. For example, Hoss features a host-access ORAM mechanism that goes beyond traditional performance constraints and incorporates several data-dependent optimizations that are not possible in prior designs. We implement a prototype of Hoss and benchmark it against Compass. Our results show that Hoss achieves up to 67x speedup while maintaining high recall, with larger gains at scale.

Karger's elegant random contraction algorithm for finding a global mincut in a graph has been highly influential. More recent work has obtained several different (nonuniform) random contraction algorithms for mincut in hypergraphs and hedgegraphs. Motivated by the conceptual goal of understanding these algorithms in a unified fashion, we study random contraction algorithms for finding a minimum quotient of a polymatroid. We introduce the notion of quotient-bounded polymatroids and show that several existing results can be derived and understood under a common algorithmic framework for quotient-bounded polymatroids.

Recent systems train coding agents with reinforcement learning (RL) on trajectories collected through multiple execution harnesses. However, gains under the training harnesses leave open whether the improvements transfer to unseen interfaces and how they depend on reward normalization. We present a controlled study that measures RL gains under fixed harnesses and compares the same models across harnesses. We train two Qwen3-8B RL variants from the same supervised fine-tuning (SFT) checkpoint using identical replay data. They differ only in whether rewards for each task are normalized within each harness (Within) or across harnesses (Cross). On SWE-bench Verified, we identify three main findings: (i) Multi-harness RL yields modest gains over the SFT baseline under both normalization rules. Across the four training harnesses, the estimated mean improvements are +0.48 and +0.77 percentage points for Within and Cross, respectively. (ii) Cross increases the fraction of trajectories with nonzero advantage but provides no consistent advantage over Within on an unseen harness across training seeds and evaluation budgets; neither rule significantly improves over SFT on this harness. (iii) Performance varies substantially more across harnesses than with RL training. After averaging over model variants, the highest and lowest source-harness solve rates differ by 7.13 percentage points, far exceeding the estimated training gains. Even after multi-harness RL, the observed performance differences across execution interfaces remain substantially larger than either the gains from training or the differences between normalization rules.

Recent work on controllable music generation has focused on autoregressive models, leaving diffusion-based systems comparatively underexplored. We present a lightweight method for steering the pitch content of audio produced by Stable Audio Open, a latent diffusion model for music synthesis. A small convolutional probe containing approximately 125k parameters is trained to decode frame-level pitch-class activations from the model's variational autoencoder latent space, using paired audio and MIDI data. At inference time, the frozen probe serves as a differentiable loss function: its gradient with respect to the denoising latent is used to nudge generation toward a user-specified pitch-class sequence, requiring no retraining or architectural modification of the base model. Across 27 evaluation trials spanning 9 text prompts and 3 target melodies, probe-guided generation increases melodic coherence by 2.4x over the unguided baseline (p < 1e-5, Wilcoxon signed-rank test), demonstrating that musically meaningful structure is both recoverable and steerable in diffusion-based music latent spaces.

Compound AI workflows are increasingly used to serve complex AI tasks by coordinating multiple AI models and software components. This approach enables deployment flexibility, as each workflow stage can expose different model variants and resource requirements, but it also expands the deployment choices. A deployment must choose an execution plan that selects AI models for each compound AI workflow stage and places them on a heterogeneous cluster in order to satisfy SLOs. Deployment optimizers therefore need estimates to compare many candidate plans and identify feasible ones. System metrics can often be profiled per stage and composed according to workflow topology, but accuracy cannot, as errors and information loss at upstream stages affect the accuracy of downstream stages. Existing approaches either profile complete configurations end to end, which scales poorly, or use product-based accuracy surrogates that treat stages as independent and can misrank candidate plans. We introduce Atlas, a framework for optimizing compound AI deployments under SLO constraints. Atlas uses MAP, a Markovian Accuracy Predictor, to estimate configuration accuracy from local conditional accuracy transitions between adjacent workflow stages. MAP discretizes intermediate outputs into accuracy buckets and composes transition profiles according to workflow topology, giving the optimizer an accuracy estimate without exhaustive end-to-end profiling. Atlas formulates execution-plan selection as a mixed-integer linear program that maximizes predicted accuracy subject to SLOs. Across four compound AI workflows, MAP achieves Spearman correlation up to 0.947 while reducing profiling cost by up to 2.6x relative to exhaustive end-to-end profiling. Guided by MAP, the Atlas optimizer selects execution plans within 0.03 of oracle accuracy while reducing deployment cost by up to 42% through heterogeneous placement.