Natural images are often conjectured to lie on a low-dimensional manifold within the high-dimensional pixel space. Recent work leverages this to build self-contained pixel-space diffusion models without specialized machinery or pretrained components. A plain Transformer on raw image patches suffices, provided it predicts the clean image rather than the noise and compresses each patch through an aggressive bottleneck. The mixing operator, however, has remained softmax attention, inherited from sequence modeling rather than derived from denoising. For Gaussian noise, the optimal denoiser admits a canonical expansion in polynomial functionals of the observation, the Wiener--Volterra series. We prove that softmax attention, and any normalized mixing rule (including linear attention), cannot realize any polynomial map beyond an affine one. In contrast, Volterra layers realize this class directly and, on a data manifold, achieve approximation rates governed by the manifold dimension rather than the ambient pixel dimension. We therefore propose the architecture these results prescribe; raw image patches passed through a bottleneck embedding, followed by cascaded low-rank Volterra layers, which carry no activation function because each layer is nonlinear in its own right. To our knowledge this is the first pixel-space diffusion architecture whose mixing operators are Volterra layers throughout, with no attention, no convolution, no pretrained components, and no activation function on the residual path. On ImageNet $256 \times 256$ and $512 \times 512$, the resulting model is competitive with self-contained pixel-space diffusion at matched parameter and FLOP budgets.
Training datasets for unsupervised anomaly detection are often assumed to be anomaly-free, but in practice, they frequently contain malicious or anomalous contamination. This contamination severely degrades the dependability and security of downstream detection models. While some contamination-robust methods exist, they rarely generalize across different model architectures and typically rely on a priori knowledge of the contamination ratio\textemdash information that is rarely available in real-world security operations. In this paper, we propose a novel, self-contained preprocessing framework for anomaly detection that identifies and removes training set contamination without relying on known contamination ratios. Our approach performs a dynamic analysis of the cosine similarity between each data point and its reconstruction during the early training stages of an auto-encoder, mapping points into a dynamic space where anomalous regions naturally exhibit lower density. By executing ``hard cleaning''\textemdash removing all identified anomalous candidates from the training pool\textemdash we systematically improve data quality. Because this cleaning occurs at the data level, the security and reliability improvements generalize seamlessly across various downstream anomaly detection architectures. Evaluating our framework across 8 distinct models and 10 security-relevant datasets, our approach achieves up to 100\% contamination reduction, yielding average improvements of 11.2\% in F1 score and 7.3\% in classification accuracy.
We propose a new mathematical formalism in which all kinds of tensors and of operations on those tensors are represented as simple function compositions. Tensors are functions from indices to values. An element-wise operation (e.g. an activation function) is a composition with a function applied to values; re-indexing (e.g. reshape, transposition) is a composition with a function applied to indices; and a reduction is a composition with both an index function, mapping each index to several indices, and a reduction operation (e.g. product, sum, max). Combined with tensor stacking, these compositions can represent most interesting kinds of tensors (e.g. dense, jagged, all kinds of sparsity patterns, matrix factorizations), as well as lazy versions of almost every operation available in modern tensor computation frameworks, including einsums between tensors of any kind. Identities between compositions can then be used to rewrite such expressions into equivalent ones, without evaluating them. For instance, the product of two sparse matrices can be rewritten so that only products between non-zero values are computed, turning a sparse problem into a dense one. Using this algebra as the intermediate representation of a tensor computation engine would shift the focus from defining operations on tensors (e.g. einsum between sparse tensors) to defining an algebra on index functions. This should reduce the codebase complexity from $\mathcal{O}(mn^2)$, with $m$ the number of binary operations between tensors and $n$ the number of tensor types, to $\mathcal{O}(m + k^2)$, with $k$ the number of types of index functions. A draft implementation of a tensor computation framework based on this algebra is publicly available.
Rapid, high-fidelity analysis of four-dimensional scanning transmission electron microscopy (4D-STEM) datasets from heterogeneous materials remains challenging because coexisting phases, orientation variants, local strain gradients, and intrinsic TEM thin-foil effects produce complex, overlapping diffraction signatures. Here, we present a machine-learning (ML) framework that combines physics-informed geometric descriptors with weighted high-dimensional embeddings to identify latent structural patterns with minimal prior knowledge. By constructing a regularized descriptor space before embedding and density-based clustering, the method improves cluster stability, suppresses noise-driven fragmentation, and reduces sensitivity to embedding hyperparameters. Spatial weighting produced the largest stability gain, increasing median cluster persistence from 0.04 to 0.55 while reducing the noise fraction from approximately 22% to 0%. Rotation weighting increased DBCV from 0.80 to 0.92, although the noise fraction rose from 0.2% to 1.0%, highlighting a trade-off between improved density separation and noise rejection. We validate the framework on two benchmark systems: Si/Si$_x$Ge$_{1-x}$ epitaxial layers with continuous lattice variation and deformation, and multiphase Ti-Nb alloys containing distinct phases and orientation variants. Comparisons with a pretrained vanilla autoencoder and model-free pattern vectorization show that the physics-informed geometric descriptor provides higher cluster fidelity, greater interpretability, and substantially lower dimensionality.
Public automatic speech recognition (ASR) benchmarks increasingly report low sequence error rates, yet production systems still use hotword lists, contextual biasing, retrieval, and post-correction. This motivates a complementary question: does a transcript preserve selected names, products, technical terms, versions, and relational phrases? We present KeyASRBench, a bilingual pilot that operationalizes occurrence-based Keyword Error Rate (KER) over frozen whole-phrase annotations. The 10-hour model-conditioned challenge set contains 1,501 Chinese and English clips with 4,723 keyword occurrences. Across five locally inferred checkpoints and two official API systems, language-specific MER/WER ranges from 5.31% to 12.91%, while KER ranges from 13.61% to 33.81%. A disjoint 30-minute model-blind control yields KER of 10.65%-27.31%. The control confirms that selected-phrase errors extend beyond challenge-set selection.
In a multi-turn conversation, the choice of language model at one turn can affect which model should answer the next. The selected model's reply shapes the context for the next query, and its cached prompt prefix can reduce the cost of using that model again. We introduce MultiTurnRoute, a platform for building and evaluating multi-turn LLM routers. A shared interface gives built-in and user-defined routers the current query, dialogue history, and an estimate of each model's reusable prefix length. Our routing algorithm, Chain-of-Route (CoR), uses a learned gate over recent and recurrent dialogue history to predict response quality and estimates cost from each model's reusable prefix. MultiTurnRoute Arena ranks routers on recorded conversations and compares session quality and cost in interactive runs where a simulated user responds to each reply. On a benchmark drawn from WildChat and LMSYS-Chat-1M with per-turn quality annotations for five LLMs, CoR reduces the normalized cost required to match the strongest model's quality by up to 20.3% relative to the strongest routing baseline, while matching or improving peak response quality.
Parcel-scale land-use data is missing for most rapidly urbanizing Global South cities, as current segmentation architectures rely on multispectral imagery and planned layouts. We introduce a hand-annotated six-class land-use corpus covering 202 km$^2$ of Kathmandu Valley across 408 fine-resolution RGB scenes. The dataset is uniquely challenging: RGB spectral similarity is high, class imbalance approaches 100:1, and key structural features like rivers measure just two to three pixels wide at a standard hierarchical decoder's prediction stride. Evaluating five architectures reveals that standard random data splitting conceals massive spatial leakage. Switching to a strict sequence-block split drops performance by up to 0.095 mIoU. This penalty exceeds the performance gap between the architectures themselves. We modified the SegFormer baseline by integrating a half-resolution residual detail path, boundary-reliability reweighting, a topology-aware loss, and a presence-balanced sampler. Adding only 1.9% more parameters, the enhanced architecture improves mean IoU across every test fold, with gains concentrated in rare, thin-structure classes. However, both models achieve identical results on the external LoveDA benchmark. Finally, augmenting with diffusion-generated pairs at matched compute degraded accuracy across all architectures, especially harming rare classes it aimed to improve. This demonstrates that blindly incorporating synthetic data can severely damage performance, highlighting a critical need to rigorously evaluate and filter generated samples before training.
Running a language model involves more than its trained weights. The same model can receive very different benchmark scores, produce different answers, or cost more to serve depending on how it is evaluated and run. In one experiment in this book, changing only the maximum response length moved accuracy on a math benchmark from 0.08% to 93.18%. This book explains how inference settings shape the results we report and the systems we deploy. We begin with evaluation, including generation limits, answer extraction, scoring methods, and quantization. We then examine why outputs can change across batches, attention backends, and caching, even when sampling is disabled. The next chapters cover serving performance, including batching, memory capacity, latency, throughput, and cost. We also study structured outputs, tool calls, reasoning budgets, and long-context tasks, showing what these controls and benchmarks do and do not establish. The book concludes with practical guidance for recording inference runs, checking evaluations, estimating capacity, and comparing systems fairly. Written for researchers and engineers working with language models, it aims to make inference results easier to understand, reproduce, and trust.
Reinforcement learning with verifiable rewards (RLVR) has become an important way to improve how language models reason, solve problems, and use tools. But getting it to work in practice is not always straightforward. Training can fail because a model receives no useful reward, learns to exploit the reward system, stops exploring, or appears to improve when the evaluation itself is misleading. In this book, we give a practical introduction to reinforcement learning post-training for language models, focusing on what works, what fails, and why. We begin with the foundations of post-training, explain GRPO and the training loop step by step, and then follow five training programs covering math reasoning, search agents, training stability, mixture-of-experts models, and software engineering tasks. Each program shows the problems encountered, the changes made, and the lessons learned. The book also examines what reinforcement learning actually changes inside a model. The remaining chapters provide a practical reference for reward design, evaluation, training tool-using agents, debugging common failures, reproducibility, rollout engines, and hardware. Written for researchers and practitioners working with open models, the book aims to help readers understand how to train models, recognize when something goes wrong, and decide what to try next.
Tree decision diagrams (TDDs), recently introduced by Capelli et al. (SAT 2026), are representations of Boolean functions. They generalize OBDDs by replacing variable orders with vtrees. The width of a TDD counts the distinct subfunctions across each vtree cut. Counting subfunctions is blind to linear structure. We introduce algebraic tree decision diagrams (ATDDs) by replacing this count with the rank of the cut matrix. The maximum of these ranks over all cuts, the factor rank, is exactly the width of the ATDD. Every TDD can be converted into an ATDD of no greater width. An ATDD is a tree tensor network in canonical form. ATDDs are exponentially more succinct than TDDs, and quasi-polynomially more succinct than structured deterministic DNNFs. Our main result is that this gain in succinctness does not cost canonicity. ATDDs are the first strict canonical form for a rank-based representation over arbitrary vtrees. For each function and vtree there is exactly one ATDD, over the rationals and without any genericity assumption. The canonicity admits a local characterization. Whether a tree tensor network is the ATDD of the Boolean function it computes can be checked in polynomial time from its tensors alone. Equivalence checking becomes a test of syntactic equality. The operation set includes polynomial-time model counting, weighted model counting, and a test whether a tree tensor network represents a Boolean function. Capelli et al. asked for the complexity of changing the vtree of a TDD and suggested local vtree changes. We answer the question. After a rotation of the vtree at one node, the new ATDD differs from the old one in two tensors only. Vtree change runs in output-polynomial time both for TDDs and for ATDDs. We formalize and verify our core results in Lean. We provide a prototype implementation of TDD and ATDD compilers as a first empirical validation.
Conflicting stakeholder objectives make it difficult to coordinate investments in urban resilience. We model conflict-driven urban design as a non-zero-sum digital board game in which stakeholder agents allocate scarce resources and balance individual payoffs against shared urban development and resilience objectives. We use multi-agent reinforcement learning (MARL) to train policies and examine their behavior across three settings. On a 4 x 4 board, learned policies reach about 84% of the total return found by a simulated annealing benchmark. On an 8 x 8 board with spatial and resource constraints, trained agents favor projects that improve resilience in self-play. In a 12 x 12 simulation initialized from historical maps of Tatsumi, Tokyo, trained agents achieve a mean total score of 452.1, compared with 357.3 for a random-policy baseline across 10,000 simulation runs. Gains in adaptability, sustainability, flood resistance, and greenery come with reductions in vitality and density. A human-in-the-loop interface supports replay and lets participants take over stakeholder decisions, providing a testbed for exploring these trade-offs in participatory urban planning.
Plot-to-code generation has been greatly advanced by recent multimodal large language models (MLLMs). However, existing studies mainly focus on one-pass generation or iterative sampling guided by verifier-based filtering, which does not align with the typical human programming workflow of generate first, then iteratively error-location and fix. This naturally motivates a test-time refinement paradigm for plot-to-code generation. In this work, we present the first study on plot-to-code generation via iterative test-time refinement. We propose a Visual-Code Diagnostics framework that enables refinement through visual feedback from source code. Specifically, we first categorize generation errors into four coarse-grained types, and then train lightweight discriminators to recognize and localize these errors based on the rendered visual outputs and corresponding code. Unlike traditional compiler- or debugger-based verification, our approach performs iterative refinement guided by one diagnostics system, enabling more effective semantic and visual correction during test time. Extensive experiments demonstrate that our method consistently improves generation quality across diverse generators and significantly outperforms strong test-time scaling baselines.
Cell types are organized by taxonomic relationships that define lineages in the so-called cell ontology. However, most existing foundation models typically ignore cell-type lineages encoded in this cell ontology. We introduce a biologically informed foundation model called scOntoFM, which embeds hierarchical cell ontology into representation learning. Using Lowest Common Ancestor (LCA) distances from the ontology graph, our framework pairs efficient offline triplet sampling with hierarchical cell-ontology learning. This approach jointly preserves local neighborhoods and shapes the global embedding geometry to reflect the ontology structure. Across diverse cell- and gene-level benchmarks, the model consistently improves performance, with especially strong gains in zero-shot settings. Our model also balances batch integration with biological structure preservation, providing complementary perspectives to existing embedding evaluation frameworks and supporting scalable biological discovery.
AI agents and reinforcement-learning (RL) trainers run thousands of short, isolated GPU jobs: tool calls, verifiers, and rollouts. A job often finishes in seconds, but starting a GPU pod for it takes 20-35 s even when the container image is already on the node, and minutes when it is not. A warm pod removes the wait only if it can serve the next job, and keeping pods warm must neither waste GPUs nor leak state between callers. Firepool, the sandbox and execution broker behind agentic and RL workloads at LinkedIn, meets both requirements. It makes pods interchangeable by building them from a shared base image per hardware profile and binding each job's code, extra dependencies, and cluster topology when the job starts. It keeps them warm cheaply and safely by reusing a pod only for the caller that used it last, staging zero-VRAM standby pods in fractional GPU shares alongside active pods, and capping the GPUs that warm pools may hold via queue-driven budgets. Over a sampled peak period in production, Firepool serves 94.7% of 3,206 pod requests from warm or standby capacity. In paired benchmarks with fixed-duration jobs on production H200 nodes, agent and RL loops finish 13x and 10.8x faster than with a fresh pod per job; against a fresh reuse tag per job, warm reuse cuts p90 start latency from 59.1 s to 3.5 s. When a job cannot reuse the previous pod, standby promotion shortens the handover wait from 51 s to 0.94 s. In local and simulated runs, a keep-alive matched to the loop's pauses keeps the latency benefit while holding about half the capacity of an always-on pool.
Vision-Language Models have had massive leaps in recent years. Yet, when tested on high-level cognitive tasks like pattern completion, abstract logic, and relational reasoning, they still fall short. This limitation stems from the models' lack of ability to find out underlying rules and maintain constant intermediate steps, rather than merely deciding on the correct option. To research these shortcomings, this study analyzes visual reasoning performance throughout a multilingual evaluation framework encompassing the B\.ILSEM, MM-IQ, Visual Puzzles, and PuzzleVQA datasets. We evaluate various data mixing strategies and supervised fine-tuning (SFT) methods. We observed that a correct answer does not always mean the reasoning chain is valid. Because of this, we adopt an LLM-as-a-Judge approach to assess the quality of reasoning beyond final answer accuracy. Our experimental results reveal that baseline models often exhibit more stable reasoning compared to fine-tuned counterparts. Most importantly, we observe that SFT can lead to model collapse, manifesting as infinite text loops or empty outputs. The study concludes that current VLMs still struggle with abstract visual reasoning.
High-level synthesis (HLS) can translate C/C++ kernels into register-transfer level (RTL) designs, yet generating high-quality hardware still relies heavily on manual pragma tuning and repeated synthesis evaluation. We propose AutoC2V, a C++-to-Verilog generation framework driven by real synthesis feedback that enables large language models (LLMs) to directly generate RTL with improved quality of results (QoR) from synthesizable C++ kernels. We construct training data from real synthesis results, including synthesizable C++ source code, the corresponding Verilog RTL, QoR metrics, and optimization objectives. Using this dataset, we align the model toward synthesis-aware RTL generation through Supervised Fine-Tuning (SFT) and Group Relative Policy Optimization (GRPO). To reduce the high cost of repeatedly invoking real synthesis during GRPO, we design the QoR Prediction Reward Model (QoR RM), which predicts latency and resource metrics by encoding RTL into long-context semantic representations, thereby providing low-cost and continuous QoR reward signals for policy optimization. Experimental results demonstrate that AutoC2V not only substantially improves RTL functional pass rates, but also achieves superior latency and resource optimization compared to the unfine-tuned base model, representative HLS optimization approaches (HLSPilot, LLM-DSE, and AutoDSE), and advanced general-purpose commercial models (Claude Opus 4.8, GPT-5.5, Gemini 3.1 Pro, DeepSeek v4 Pro, and GLM-5.2). These results indicate that combining real synthesis data with efficient QoR reward modeling can further improve the ability of LLMs to generate high-quality RTL.
Iterative neural policies usually spend the same computation on every decision, although the benefit of additional processing varies across states. We introduce Adaptive Internal Computation (AIC), an iterative policy that learns when to stop refining its latent representation and act. In AIC, we train action selection and stopping jointly, an expected-depth penalty values computation and intermediate value predictions provide auxiliary supervision. We show that when only some states benefit from further refinement, an ideal state-dependent allocator outperforms every state-independent stopping schedule at the same expected depth. With at most eight refinement steps per decision, AIC executes 3.3 steps on average across twelve benchmark tasks and lowers per-decision latency by 31--44% relative to a fixed eight-step policy on four control and robotics tasks. Its mean reward exceeds the best fixed-depth policy on seven tasks and trails it on four. Ablating adaptive stopping and value supervision separately shows that stopping accounts for most of the reward gain on PickCube and AntMaze, where it also executes more refinement steps and that supervision accounts for most of it on RareEvent.
Recent progress in vision foundation models (VFMs) has made automated open-vocabulary 3D bounding box annotation increasingly feasible by jointly leveraging LiDAR point clouds and images. However, partial observations remain a major challenge in sparse driving scenes, where visible LiDAR points often cover only a small fraction of the object's surface and provide insufficient cues for estimating its full spatial extent. As a result, existing methods often produce mislocalized boxes or leave objects unlabeled. To address this limitation, we propose \textbf{\Ours{}}, an automated annotation framework that leverages image-generated 3D assets as amodal geometric priors for partially observed objects. By establishing correspondences between input images and rendered views of 3D assets, \Ours{} robustly aligns canonical amodal geometry with partially observed world-space LiDAR point clouds. To improve temporal consistency and reduce computational cost, \Ours{} generates a single asset per instance, estimates its metric scale once at a single frame and propagates it across frames for dynamic instances, and aggregates multi-frame observations for static instances. 3D object detectors trained on annotations from \Ours{} improve performance over strong baselines by 14.1 mAP on the nuScenes dataset and by 12.45 AP at IoU 0.5 for the vehicle class on the Waymo Open Dataset (WOD). Our project page is available at: \href{https://oliver0922.github.io/AmodalBox/}{https://oliver0922.github.io/AmodalBox/}
Real-time cooperative perception (CP) in multi-agent traffic environments is highly resource-intensive, which posing scalability challenges for large-scale deployment. Efficiency can be significantly improved by selectively performing CP only in safety-critical scenarios and for key occluding objects. Vision-Language Models (VLMs) are well-suited for complex reasoning in dynamic and uncertain environments, making them a promising tool for this selective CP task. This study proposes a VLM-driven agentic framework for blind spot reasoning and dialogue-based CP. A mixed-model architecture is designed that integrates conventional object detection for stable perception, vision-language reasoning for identifying impactful occlusions, and a physics-based geometric module to align reasoning with real-world visibility constraints. The framework also incorporates a Chain-of-Thought (CoT) prompting mechanism to enhance interpretability and intent-aware decision-making. We collect real-world data from connected vehicles operating in urban traffic and compare the driving-critical blind spot identification and dialogue performance of four state-of-the-art VLMs. Experimental results demonstrate that our method achieves over 80% accuracy in driving-critical blind spot detection and over 90% success in dialogue-driven perception sharing. This work provides a scalable, interpretable, and bandwidth-efficient alternative to conventional CP pipelines, and offers new insights into agentic communication design for future connected traffic systems.
In this work we proposed a architecture designed for the classification of ASD in toddlers using eye-tracking data, addressing the critical need for early and precise detection. The proposed model incorporates FMMix, a feature-level augmentation technique that bolsters the model's learning capacity, and in this work we proposed the ATHENA, which captures intricate feature relationships via an adaptive hypergraph framework. Evaluated on two datasets dataset of gaze coordinates-the model achieves state-of-the-art performance, with AUC scores of 99.80% and 99.86%, respectively. It also exhibits high accuracy of 98.45% and 98.56%, sensitivity 98.01% and 98.10%, and specificity 98.74% and 99.02%, respectively. Demonstrating its superior ability to identify subtle gaze patterns associated with ASD, these results significantly surpass existing methods, highlighting the model's potential for clinical ASD screening. Future efforts will focus on validation across broader populations and integration with multi-modal data to further refine diagnostic accuracy.