让下一篇重要论文，不再错过。

Online 3D bin packing with a processing buffer requires choosing which box to pack, its orientation, and its placement while keeping the stack stable under unknown future arrivals. A collapse ends the episode and forfeits the remaining volume. Existing methods use geometric rules or hard stability masks, limiting how risk can influence the joint choice of box, orientation, and placement. We propose Stability-Hybrid On-Policy Reinforcement Learning (SHORE), which combines a learned stability probability of a placement with an on-policy estimate of its return conditional on stability. A 3D U-Net learns stability from executed outcomes and simulated alternatives, revealing outcomes of actions underrepresented in early rollouts. A discrete on-policy actor--critic uses stability predictions in both its action values and policy prior, with gates that limit the effect of an untrained or changing model. In deployment, SHORE does not require additional search or simulation overhead. In Isaac Gym, SHORE reaches 73\% space utilization by 0.4M environment steps, while the same learner without stability modeling reaches that level at about 1M. On the expanded 60-box set, SHORE achieves 0.734 space utilization versus 0.588 for the GOPT state-of-the-art baseline, with a higher placement-failure frequency (1.03\% vs. 0.63\%), trading more risk for greater packed volume. The results show faster learning and higher utilization on the more demanding realistic box arrival set. Ablations suggest that simulated alternatives provide useful early supervision, while executed outcomes alone sustain later performance. We also demonstrate the method on a real robot to verify the simplicity of integrating the method to real bin packing pipelines.

Quantization reduces the memory and bandwidth demands of large language model (LLM) inference but can degrade accuracy; offloading higher-precision data can instead stall generation. We introduce StrataFlux, a framework that challenges this tradeoff by recovering accuracy lost to low-bit weight and key-value (KV) cache quantization while retaining quantization's memory and speed advantages. StrataFlux co-designs refinement representation, selection, and delivery to make host RAM and solid-state drives useful sources of additional precision. Its selective precision streaming reduces transfer volume, overlaps delivery with computation, and reuses fetched precision across decoding steps so generation need not wait for every transfer. We evaluate 2-4-bit weights and 2-3-bit KV-cache on dedicated GPUs and unified-memory platforms. For 2-bit Llama-3.1-8B-Instruct weights, StrataFlux reduces WikiText-2 perplexity from 9.46 to 8.40 while maintaining 95.4% of the quantized base's throughput on RTX 5090. With 2-bit weights and KV, StrataFlux raises Qwen3-8B's AIME 2024/2025 accuracy from 11.0% to 47.1% at unchanged GPU memory use. At 128K context, Llama-3.1-8B-Instruct with 3-bit weights and 2-bit KV achieves comparable perplexity and speed to vLLM's 4-bit AWQ weights and 8-bit KV on RTX 5090, using approximately half the GPU memory. These results show that GPU-resident precision need not set the accuracy ceiling, nor offloading-tier latency dictate decoding speed.

At the edge of stability, the iterates of gradient descent oscillate along the sharp directions of the Hessian. This poses a significant challenge for continuous-time models as the usual continuous-time approximation -- the gradient flow -- admits no oscillations. Several continuous-time models have been proposed to capture the behavior at the edge of stability, among them central flow and rod flow. Although both models can be interpreted as performing a coarse-graining in time, the connection between them remains poorly understood. Here we show that both flows can be derived from a third object, the edge coupling: a functional on consecutive iterates whose critical points are the fixed points and period-two orbits of gradient descent. It is revealed that the two flows correspond to coarse-grainings at different time scales. Rod flow coarse-grains over only the fastest time scale, while central flow coarse-grains over both the fast and intermediate time scales. Furthermore, using the edge coupling, we derive rod flows and central flows for momentum methods such as heavy ball and Adam. We verify the derived flows experimentally on representative neural network architectures, where they track the discrete iterates throughout all phases of training -- including the transition from the stable regime to the edge of stability.

In data clustering, it is often desirable to find not just a single partition into clusters but a sequence of partitions that describes the data at different scales (or levels of coarseness). A natural problem then is to analyse and compare, as a whole, the (not necessarily hierarchical) sequences of partitions that underpin such multiscale descriptions. Using tools from topological data analysis, we introduce here the Multiscale Clustering Filtration (MCF), a filtration of abstract simplicial complexes that encodes arbitrary cluster assignments in a sequence of partitions across scales of increasing coarseness. We prove the stability of the MCF with respect to small changes in the sequence of partitions, and establish that its persistence diagram (PD) provides an interpretable representation of the multiscale clustering: at dimension 0, it captures violations of the refinement order, and at higher dimensions, it tracks the emergence and resolution of higher-order inconsistencies between cluster assignments. We also introduce an equivalent construction of the MCF via a nerve complex filtration, and show that this nerve-based MCF has computational advantages in certain settings, aiding scalability. Further, we show that in the hierarchical case, the MCF reduces to a Vietoris-Rips filtration of an ultrametric space, and we establish that pairwise measures, such as conditional entropy or zigzag persistence, are fully determined by a graph representation of the Sankey diagram and thus fail to describe the higher-order inconsistencies across scales that MCF captures. Using synthetic and real-world data from stochastic block models, single-cell transcriptomics and temporal social interactions, we illustrate how the MCF PD serves to characterise and compare multiscale clusterings beyond pairwise or hierarchical baseline methods.

Background and Objective: Our main goal is ultrasound guidance: moving a probe until its image matches a target view. Published methods for each part of a guide are evaluated in separate simulators. We asked whether these methods, combined into one guide, are good enough. Methods: We built USB Gym, a testbed whose standard engine simulates ultrasound from public patient CT, with a network drawing organ textures from CT labels. Its guide has three replaceable parts (slots), an organ segmenter, a controller and a reward, which received 20 published methods, each run on the same 60 episodes of six held-out patients. In each slot, a published method replaced ours only if detectably better on these patients. The better of this combination and our original guide on 20 new CTs is the benchmark baseline. A pose oracle, reading the exact probe pose available only in simulation, set the reference. The results below are exploratory. Results: On these CTs, the probe started on average 16.0 mm from the target view; the pose oracle brought it to 1.5 mm. Of the results known when the combination was fixed, only an image-similarity reward (SSIM) was slightly but detectably better, so the combination was our guide with that reward. This combination was the benchmark baseline, ending at 13.7 mm (16% of the oracle's progress). Conclusions: Combining untuned published methods with ours did not guide the probe near the target view; the task remains under-explored. The testbed accepts new methods; no guide was tested on hardware.

Millimeter-wave radar resolves the micrometer displacements of a moving body as micro-Doppler, enabling gait, gesture, activity, pose and vital-sign sensing. Each task relies on a downstream model trained on real data, yet radar datasets lag orders of magnitude behind vision and language. Brute-force collection will not close that gap, and existing synthesis methods (forward-only simulators, learned generators and static-scene inverse renderers) cannot fit a real capture of a moving body, leaving a sim2real gap. We present 4D Point Splatting (4DPS), a physics-based differentiable renderer that fits an explicit moving scene to a real capture. Using the synchronized camera to initialize the room and body, we represent the scene as oriented points carrying material, normal and velocity, and splat each point into range and Doppler bins. Our output is the raw complex return a real radar records, so we apply the same signal processing on the rendered data to obtain any representation for training any SOTA downstream model. We fit 235 sequences on HuPR and 100 on RT-Pose, two datasets with different radars. Our reconstructions set the state of the art, with median correlations of 0.940 and 0.830 on range-azimuth, 0.829 and 0.695 on range-Doppler, and 0.988 and 0.966 on micro-Doppler. Our renders also set the state of the art in downstream sim2real gap. On RT-Pose, a pose model trained on them reaches 21.9 cm MPJPE on real data, against 16.7 cm for real training and 127.6 cm for the RF-Genesis baseline, where both the rendered- and real-trained models learn nearly the same features. Our custom CUDA kernels fit a sequence in 5.7 minutes on HuPR and 12.9 on RT-Pose on one H100, making dataset-scale inverse rendering of mmWave radar possible for the first time. This work can enable synthetic data generation for any human-centered downstream task and model, and scene editing for data augmentation.

We study learning rate (LR) scaling for hybrid architectures combining Transformer and State-Space Model (SSM) blocks, a class adopted by several recent production language models. In particular, we focus on the gap between the theoretical scaling rules derived for SSMs under zero-order-hold (ZOH) discretization at infinite width with growing state size, and the field-standard practical implementations using simplified-ZOH Mamba at fixed state size. Surprisingly, in this practical regime hybrid architectures achieve a near-zero LR transfer gap across widths 256-2048 and depths 4-32 up to billion-parameter scale using only the original $μ$P prescription, even though SSM operations fall outside its Tensor Programs representability conditions and every parameterization we test fails the standard coordinate-check diagnostic of $μ$P correctness. We attribute this to a two-condition decomposition of LR transfer in hybrid architectures: a global update-to-weight invariance, enforced by $μ$P's initialization and LR scaling; and a local per-component balance, provided by AdamW's per-parameter normalization. Our observations show that the optimal LR is invariant to width up to 8$\times$, that this width invariance holds across depth, sequence length, batch size, and Transformer-to-SSM ratio, and that it transfers to Nemotron-H, a production hybrid outside our custom architecture set. We hope these findings fill the gap between theoretical scaling rules and practical hybrid implementations, and stimulate further research toward bridging it.

Fine-tuning large language models on narrow data with harmful content produces broadly misaligned behavior on unrelated prompts, a phenomenon known as emergent misalignment. We propose that emergent misalignment involves persona-model collapse: a deterioration of the internal representations that encode and distinguish personas, undermining the nuance and consistency of their expression. We first test this hypothesis behaviorally, by comparing how models fine-tuned on harmful data and benign controls answer questionnaires while role-playing diverse personas. Across four model families, harmful fine-tuning produces more polarized, all-or-nothing answers, suggesting more binary, less nuanced persona expression. Histograms show increased concentration at the scale endpoints, and Kullback--Leibler (KL) divergence quantifies the distributional shift. We also find reduced within-persona consistency: repeated answers to the same question under the same persona become less stable. We reproduce these observations across diverse harmful datasets and questionnaires that probe different aspects of persona-conditioned behavior, while capability and simple instruction-following controls argue against a uniform loss of general ability. Finally, a mechanistic analysis of persona-conditioned activations in Qwen3-235B shows that representations elicited by different persona prompts span fewer effective dimensions in the misaligned variant than in the original and benign-control variants, most clearly in the final layers. Together, these findings provide behavioral and representational support for persona-model collapse as an aspect of emergent misalignment.

Data engineering for large language model (LLM) reinforcement learning (RL) decides which samples to train on, when to present them, and which to batch together. Existing methods mostly rely on external feedback, such as pass rates or difficulty labels, which observes only the outcome on each sample. We argue that model internals offer richer and more model-intrinsic feedback that can guide all three decisions. We propose SAERL which organizes RL training data in the feature space of Sparse Autoencoders (SAEs), a mechanistic interpretability tool that decomposes model internals into sparse, interpretable features. SAE features let us view the model as a set of functional regions, each formed by features that related samples activate together. For each sample, its SAE activations indicate which regions it engages, and its gradients projected onto SAE features indicate how training on it would change these regions. Guided by this view, SAERL keeps samples with strong and non-redundant gradient signals, trains each functional region from easy to hard, and trains one functional region at a time, drawing each batch from a single activation cluster with moderate cross-cluster mixing. The resulting curriculum is constructed once with small models and transfers to larger ones. Experiments on mathematical reasoning show consistent gains across model scales and RL algorithms. For instance, on Qwen2.5-Math-1.5B, SAERL improves the average accuracy of GRPO by 3.1% and reaches the target accuracy with 42% fewer training steps. Further analyses reveal that ordering and batching are complementary. These results demonstrate that model internals are a powerful and practical foundation for post-training data engineering.

While Diffusion Language Models (DLMs) are theoretically well-suited for iterative refinement due to their non-causal structure, off-the-shelf masked diffusion language models (MDLMs) often fail to reliably revise incorrect tokens in practice. The key challenge is that the model cannot tell correct from erroneous visible tokens: standard MDLM training supervises only masked positions, undermining confidence-guided refinement. We study corrective behavior in DLMs, the ability to assign lower confidence to incorrect tokens and iteratively refine them while preserving correct content. Under matched compute and data, we show that conventional masked-diffusion fine-tuning induces much less of this capability than a correction-oriented post-training principle that explicitly supervises visible corrupted tokens, which enables discriminative confidence and targeted refinement. We further find that also supervising clean visible tokens, as in BERT- or GIDD-style training, stops remask-based refinement from improving with additional steps. To evaluate it, we introduce the Code Revision Benchmark, a controllable, executable benchmark for error localization and in-place correction. The effect transfers from a 0.5B model to LLaDA-8B and persists, with a smaller margin, when corruptions are drawn from the base model's own most plausible predictions. The same error-aware confidence also makes parallel decoding more effective. In HumanEval(+) code completion at 0.5B, mathematical reasoning at 16B, and high-entropy parallel decoding with remasking (ParallelBench), models trained with our objective outperform standard masked-diffusion fine-tuning under matched compute, and outperform the base model on HumanEval(+), MATH, and the high-entropy ParallelBench average; at 8B, neither objective improves from-scratch generation over the base model. Code and data: https://github.com/zhangshuibai/CDLM

Anomaly detection in multivariate time series is a critical task across a wide range of real-world applications, where abnormal behaviour is rare, labels are unavailable, and the cost of a miss is high. The central challenge is learning a characterisation of normality precise enough to flag deviations. Representation self-supervised learning, typically through contrastive approaches, addresses this by embedding temporal patches into a latent space where normality occupies a well-defined region, with anomalies detected by geometric deviation. However, contrastive approaches shape this space indirectly through pair-sampling heuristics, providing no explicit control over the geometric structure on which distance-based scoring operates. That geometry is thus left to chance. We present VACE (Velocity-Aligned Channel Embeddings), a self-supervised anomaly detection method that represents normality as a locally smooth, directionally coherent trajectory in the embedding space. To this end, VACE trains a channel-aware encoder through a velocity-consistency objective, with no negatives and no synthetic anomalies, so that normal trajectories are locally smooth and aligned. At test time, a Mahalanobis positional score and a directional score computed against a bank of normal directions are combined multiplicatively, the directional score acting as a gate that amplifies the positional one where the local dynamics are atypical. Despite its simplicity, VACE achieves state-of-the-art performance on TSB-AD-M under rigorous evaluation, significantly outperforming more complex methods trained on substantially larger budgets.

Large language models (LLMs) are increasingly used in biomedical and cognitive neuroscience, but their ability to reason reliably over heterogeneous knowledge graphs (KGs) remains unclear. We benchmark five LLMs (GPT-5.1, Gemini 2.0 Pro, Llama 3, Qwen, and BioGPT) across five KGs: DrugBank, NeuroSynth, Cognitive Atlas, Brain Map, and GDPKG. We introduce a multitier query dataset to evaluate reasoning depth and factual accuracy under KG-aware (graph-grounded) and KG-agnostic conditions. Inter-annotator agreement was substantial, with Cohen's kappa of 0.781 (range: 0.753-0.807) and Fleiss' kappa of 0.77. Results show that general-purpose models, particularly GPT-5.1 and Gemini 2.0 Pro, generally outperform domain-specific models on multi-hop and high-level reasoning tasks, although performance declines as query complexity increases. KG grounding reduces hallucination rates across all evaluated models. Contrary to our initial hypothesis, domain-specific models do not consistently outperform general-purpose models on specialized tasks without graph grounding. These findings highlight the value of combining broad language-model capabilities with structured domain knowledge. We provide practical guidance for selecting LLM-KG integration strategies based on task complexity and computational constraints, and discuss hybrid neurosymbolic architectures for personalized medicine, cognitive disorder diagnosis, and explainable AI.

Models of collective motion usually reduce social interactions to pairwise forces, despite empirical evidence for many-body effects. We introduce the directional atomic cluster expansion (dACE), which adapts the atomic cluster expansion (ACE) of interatomic potentials to self-propelled agents. dACE provides a systematically improvable, exactly equivariant expansion of an agent's acceleration that splits uniquely into 1-, 2-, 3-body and higher contributions, along with a self term for the agent's own past motion that is fitted so that it does not absorb the interactions. On simulated collectives in two and three dimensions with known rules, dACE recovers the interaction law body order by body order from little data, while the neural model SocialNet, the current closest competitor, needs more data, has much larger errors on many-body or chiral rules and is restricted to 2D. On real data sets of tracked zebrafish and golden shiners, the laws learned by dACE agree better between disjoint halves of the data than SocialNet's. The learned laws are interpretable, and in the shiners the 3-body term opposes the pairwise response as neighbourhoods get dense. Finally, we show that next-step prediction accuracy mostly measures self-propulsion and noise rather than the interactions, and suggest more robust combinations of error measures for interaction learning.

Modern machine learning models are deployed in diverse, non-stationary environments where they must continually adapt to new tasks and evolving knowledge. Continual fine-tuning and in-context learning are costly and brittle, whereas neural memory methods promise lightweight updates with minimal forgetting. However, existing neural memory models typically assume a single fixed objective and homogeneous information streams, leaving users with no control over what the model remembers or ignores over time. To address this challenge, we propose a generalized neural memory system that performs flexible updates based on learning instructions specified in natural language. Our approach enables adaptive agents to learn selectively from heterogeneous information sources, supporting settings, such as healthcare and customer service, where fixed-objective memory updates are insufficient.

Reward machines (RMs) have been proposed as a reward formulation in reinforcement learning (RL) to capture non-Markovian tasks where credit assignment depends on history. By representing reward functions as automata, RMs expose the task's temporal structure and allow such rewards to be handled within a Markovian framework. However, they assume a labelling function, a proposition-grounding oracle for observations, which must be manually engineered for each environment, limiting their wider applicability. In this work, we address this limitation by proposing RM-VLM, a novel approach in which a large language model (LLM) generates an RM from a natural-language task description, and a vision-language model (VLM) acts as the labelling function, grounding the RM's propositions over image-based observations during RM-guided policy learning. A VLM-based value function estimator is used to assess the agent's progress towards each RM transition and derive a dense reward signal. We evaluate RM-VLM on three domains covering non-Markovian grid-world navigation, robotic manipulation, and 3D object collection. Our experimental results show that RM-VLM consistently achieves higher task success rates than existing VLM-based reward methods across all tested domains.

Diffusion language models can write many tokens per step, but must choose which tokens are safe to commit together. Mean-field decoding penalizes pairs of positions whose predictions look alike. We replace this proxy with the coupling itself: how much committing one token changes the model's confidence at another. Cheap probes inside the main forward pass measure it, with no training or extra model, and it plugs into the same mean-field rule. On three models and four benchmarks, our method needs far fewer steps than mean-field decoding and is faster in most settings, also with a key-value cache, at an average accuracy cost of 1.1 points. Shuffling the couplings or replacing them with a uniform bonus loses this gain.

Online Convex Optimization with memory and time-varying constraints (COCO-M) is a recently-introduced framework that captures the dynamics of stateful online learning and non-stochastic control with budget constraints. Despite its expressive power, COCO-M remains largely understudied. In this work, we tackle this problem in its full scale, providing an optimistic algorithm that incorporates untrusted predictions to tame universal dynamic regret and cumulative constraint violation. Our analysis leverages a potent reduction from memory to delay and treats state-dependent gradients as delayed feedback. We establish the first bounds that adapt to both environment non-stationarity (path length $P_T$) and prediction accuracy, while lifting previous restrictive assumptions such as functional separability. En route to these results, we derive findings of independent interest for the memory-to-delay reduction, and for delayed OCO with time-varying constraints. The proposed framework ensures sublinear worst-case guarantees that improve with prediction accuracy, reaching $\mathcal O(1)$ regret under perfect predictions, with cumulative constraint violation (CCV) $\mathcal O(\sqrt{T})$ for unknown $P_T$ and $\mathcal O(\log T)$ for known or online observable $P_T$ , while recovering the optimal COCO rates when memory disappears.

Generative models propose crystalline materials by learning composition and geometry, but unconditional sampling gives no direct control over the space group, which constrains whether responses such as piezoelectricity are symmetry-allowed. We present Orbitol, a crystal generator built on the Crystalite diffusion Transformer that replaces atom tokens with Wyckoff-orbit tokens. Given a space group and a collection of occupied Wyckoff positions, Orbitol generates the atom type and free coordinates of each orbit together with a compatible lattice, so every valid structure is invariant under the operations of the requested group. It generates structures two to three orders of magnitude faster than the symmetry-aware baselines at the evaluated settings. On MP-20, 13.8% of Orbitol's attempts yielded stable or metastable, unique and novel structures that retained the requested space group after relaxation, compared with 10.8% for SGFM, the next best model. We also introduce SPARC, a score that combines retention of the raw detected space group with agreement between the retained-group and training distributions. Finally, we used Orbitol to generate candidates in piezoelectric crystal classes. Of the structures selected by internal screening, 89% passed our density functional theory checks, and 79% also matched no Materials Project or Alexandria entry with the same space group. Their median figure of merit was nine times that of the piezoelectric materials computed in the Materials Project, and 24% of the new candidates ranked in that set's top 10%.

Language models must now generalize out of the box to novel environments and work inside inference-scaling search procedures, such as AlphaEvolve, that select rollouts with a variety of task-specific reward functions. Unfortunately, the standard paradigm of LLM post-training optimizes a pre-specified scalar reward, often leading current LLMs to produce low-entropy response distributions and thus to struggle at displaying the diversity that inference-time search will require. We propose Vector Policy Optimization (VPO), an RL algorithm that explicitly trains policies to anticipate diverse downstream reward functions and to produce diverse solutions. VPO exploits that rewards are often vector-valued in practice, like per-test-case correctness in code generation or, say, multiple different user personas or reward models. VPO combines multi-answer prompting with a set-level reward to train the LLM to produce sets of solutions that specialize to different trade-offs in the vector reward space. Across four tasks, VPO matches or beats the strongest scalar RL baselines on test-time search (e.g. pass@k and best@k), with the gap widening as the search budget grows. For evolutionary search, VPO models unlock problems that GRPO models cannot solve at all. As test-time search becomes standard, diversity may need to become the default post-training objective.

Health-misinformation analysis presents a dual challenge: misleading claims can affect public well-being, while systems designed to detect them may expose sensitive user-submitted text. This motivates privacy-preserving inference for public-health applications in which user queries may contain confidential medical information and trained models and reference databases are maintained on organizational servers. We design a privacy-preserving text inference framework based on CKKS homomorphic encryption for health-misinformation classification and semantic retrieval. The classification pipeline fuses BioBERT embeddings, PCA-reduced MPNet embeddings, and SVD-reduced TF-IDF features, while the retrieval pipeline combines MPNet embeddings with ISCC-SCT semantic fingerprints through late fusion. We evaluate predictive performance and benchmark plaintext versus ciphertext-based execution. The hybrid classifier achieved an accuracy of 0.971 and a macro-F1 score of 0.950, with identical reported metrics under plaintext and ciphertext-based inference. The hybrid late-fusion similarity model achieved a Pearson correlation of 0.8691 with human sentence-pair similarity scores on STS-B. For ciphertext-based retrieval, we evaluated naive and optimized implementations. The optimized implementation uses CKKS single instruction, multiple data (SIMD) packing to compute and return multiple similarity scores together, reducing single-query latency over 7,128 indexed items from 222.028 to 5.168 seconds and communication overhead from 6.11 GB to 2.13 MB. These findings demonstrate the technical feasibility of privacy-preserving text inference for health-misinformation applications and characterize its predictive utility and efficiency trade-offs.

Robot-mediated rehabilitation has the potential to deliver high-intensity, individualized therapy, but existing systems (which have not consistently improved outcomes) typically prescribe kinematic trajectories rather than explicitly targeting the neuromotor activations thought to support recovery. To address this gap, and ultimately enable activation-optimizing rehabilitation robot behavior, we present a personalizable transformer-based model mapping individuals' muscle activations (as measured via surface electromyography) to prescribed contact force trajectories when interacting with an end-effector rehabilitation robot. The model uses a transformer encoder-decoder leveraging rotary positional embeddings (RoPE) to generate a latent representation of trajectory-relevant and user-specific neuromotor factors, and augments trajectory prediction with flow matching to produce a distribution over plausible prescriptions biased toward the training distribution. Using data from 16 healthy participants completing 15 isometric trajectory tracking tasks, we evaluate generalization across held-out tasks and users, as well as the feasibility of user-specific fine-tuning within a therapy-session timescale. The model achieves mean trajectory prediction errors of <8% on held out tasks and users, with effectively instantaneous deterministic inference and sub-second sampling of trajectory distributions. An initial test using the model during live rehabilitation tasks demonstrates that, on average, participants reach the target neuromotor activations with just 15% error accumulated throughout the task. These results provide a proof of concept for data-driven, activation-targeted trajectory prescription in robot-mediated rehabilitation and establish a foundation for future validation in clinical populations. Data and code have been open-sourced as part of the OpenRobotRehab project on SimTK.

We develop and analyze multilevel gradient-regularized Newton methods for unconstrained convex optimization with Lipschitz-continuous Hessians. By computing Newton corrections in lower-dimensional subspaces, the proposed methods aim to reduce computational costs while preserving the global $\mathcal{O}\left(k^{-2}\right)$ convergence rate in function values of the full-space method. To establish this rate for multilevel methods, we introduce a class of row-nested restriction operators that ensures monotonicity of the reduced-gradient norms across levels. Numerical experiments with row-nested operators demonstrate convergence behavior comparable to full-space gradient-regularized Newton, while offering substantial computational savings on large-scale instances.

LLM pruning must preserve quality while meeting hardware- and workload-specific constraints on latency, throughput, and memory. Existing pruning workflows often target parameter counts, requiring practitioners to manually select model sizes that meet latency, throughput, and GPU memory limits. However, parameter count alone does not predict these metrics, which depend on the model architecture, workload, hardware, and serving engine. We introduce AutoCompress, a hardware-guided compression framework that directly incorporates such constraints into differentiable architecture search. AutoCompress jointly optimizes global and per-layer structure to produce heterogeneous architectures, guided by serving-engine cost models validated against end-to-end latency, throughput, and memory measurements. We evaluate AutoCompress on long-context local coding, batched datacenter serving, and latency-sensitive in-game interaction. Compared to the state-of-the-art Minitron pruning framework, AutoCompress improves average accuracy across eight benchmarks by 6.57, 2.35, and 0.57 percentage points on RTX 5070, RTX 5080, and H100, respectively. AutoCompress also compresses Nemotron 3.5 Super from 120B to 74B parameters for single-B200 BF16 serving at 128K context, retaining 95% of its mean benchmark score.

Adaptive rounding methods such as GPTQ quantize the weights of a large language model (LLM) one at a time, propagating each rounding error to the weights not yet rounded. YAQA lowers the KL divergence from the original model with two-sided adaptive rounding under a Kronecker-factored Hessian, but its error feedback takes quartic work in the size of a square layer. We present GPTQ-2D, which provably computes the same rounding with cubic work, while still rounding the entries of each anti-diagonal in parallel. Its GPU implementation uses two levels of blocking to improve hardware utilization. We also extend the algorithm to Hessians with any number of Kronecker factors. In our experiments, GPTQ-2D is as accurate as YAQA and cuts the time to quantize Llama-3.1-70B-Instruct on a single NVIDIA RTX A6000 GPU from 8.6 days to 1.6 hours. Source code is available at https://github.com/IST-DASLab/GPTQ-2D.

We investigate the problem of scaling safe reinforcement learning (safe RL) to massively parallel training regimes, where a large number of environments are exe- cuted with short rollout horizons. Although this setting improves data throughput and wall-clock efficiency, it creates a mismatch with constrained Markov decision processes (CMDPs), whose safety constraints are defined over full-episode cost returns. We address this mismatch by analyzing staggered environment resets as a phase-mixture distribution over short trajectory segments. This analysis charac- terizes the bias induced by phase coverage and motivates a phase-aggregated cost estimator that reconstructs episode-level cost estimates from short rollout batches. We further incorporate finite-sample uncertainty and distribution mismatch into a constraint-tightening framework, yielding a scalable Safe RL method compatible with on-policy optimization. Using an MJX-based implementation of Safety Gym, our experiments show that the proposed method achieves performance comparable to non-massively parallel implementations while substantially reducing wall-clock training time.

Transformers use the same forward computation stream to both predict the next token and store useful state for future token predictions. We formulate the state-prediction separation hypothesis: disentangling the two roles yields better language modeling performance. We design a Transformer variant that uses separate computation streams to separate the two functions, and conduct pretraining experiments across various scales. Our experiments show that state-prediction separation consistently offers better data- and compute-efficiency, improving validation loss and outperforming standard Transformers by 2-3 percentage points on average on downstream tasks. We also conduct extensive empirical analysis that rules out potential confounders and demonstrates the fundamental difference in the gradients our design entails.

Generalist Anomaly Detection (GAD) aims to detect anomalies in unseen domains using only a few reference samples. Recent normal-abnormal guided methods improve this setting by exploiting abnormal references, but their residuals often mix transferable anomaly cues with domain-specific appearance factors such as texture, geometry, and acquisition conditions. This residual domain entanglement limits cross-domain transfer and may produce spurious activations on novel domains. We propose DART, a plug-in framework for normal-abnormal guided GAD that learns to separate residuals into anomaly-relevant and domain-specific components. DART combines a Residual Disentanglement Module, which preserves anomaly-discriminative information while suppressing domain information, with Cross-Domain Meta-Training, which forms episodes across source domains to make the separation effective. The method is architecture-agnostic and can be inserted between residual computation and the detection head of existing GAD pipelines. Experiments on MVTec AD, VisA, and BraTS show consistent improvements over multiple base methods, with the largest gains under stronger domain shifts. https://github.com/aqeeelmirza/DART