Papers

Diffusion models represent one of the most advanced paradigms in generative modeling. Leveraging their development, a growing number of style transfer methods based on diffusion models have been proposed. However, among these methods, multi-image style transfer approaches that require at least five to ten style examples tend to achieve more satisfactory results. Single-image methods, by contrast, often struggle with either insufficient content preservation or inadequate style fidelity. This greatly limits style extraction from scarce artworks and undermines their artistic value. To address this, we propose Abstract-LoRA, a method that pushes the boundaries of single-image style transfer through lightweight LoRA training on specific U-Net blocks in diffusion models. Specifically, our work is inspired by B-LoRA, a style transfer method that achieves basic style-content disentanglement by training specific U-Net blocks. However, it suffers from a critical limitation: the inability to capture complex backgrounds. Building upon B-LoRA, our method conducts a more refined analysis of U-Net blocks, employing additional U-Net blocks and clustering-based abstraction of style images to better disentangle and balance style and content. Extensive experiments demonstrate that our proposed method not only generates visually more harmonious and satisfying artistic images but also quantitatively improves the preservation of both style and content in the final outputs.

Large language model (LLM) sustainability is increasingly a serving-systems problem, not only a training problem. In production, energy and carbon impact depend on more than model size: workload shape, batching, key-value (KV) cache reuse, prefill/decode placement, model and accelerator choice, power state, geographic carbon intensity, and service-level objectives (SLOs) all matter. Recent systems papers study many of these factors separately. This paper connects those results and asks a practical engineering question: what do they imply when the decision point is a distributed inference control plane such as llm-d? The contribution here is synthesis, not a new set of benchmark results. Reported performance, energy, carbon, and cost improvements remain the results of the cited papers and systems. I group the literature into recurring design patterns and use those patterns to sketch a Sustainable Inference Control Plane (SICP) for llm-d. The proposed control plane would consider latency, energy, carbon, cache reuse, serving cost, and quality when routing and scaling, while keeping TTFT/TPOT SLOs as hard constraints. I also outline an evaluation framework based on SLO-satisfied goodput per joule and per gram CO2e, together with a reproducible experimental plan. The main observation from connecting the literature is that sustainable LLM inference is unlikely to come from one "green" model or one accelerator; it is more naturally treated as a control problem across model, phase, cache, hardware, replica, region, and time.

Rollcast is a probabilistic forecasting method for univariate time series that combines a compact set of rolling statistical anchors rather than relying on a single global model. Rolling means, medians, extrema, regression endpoints, and quantiles define candidate forecast locations and a representation of the current state. A state-dependent softmax gate learns anchor probabilities by minimizing negative log predictive density, while residual distributions retrieved from similar historical states provide local uncertainty. Recursive simulation propagates the resulting mixture over multiple forecast horizons. The method is evaluated in a Monte Carlo study covering eight data-generating processes, including autoregressive, random-walk, local-trend, threshold, regime-switching, stochastic-volatility, heavy-tailed, and variance-break dynamics. Across 2,000 independent fitted series, Rollcast is compared with the true conditional predictive distribution generated by an oracle simulator. Overall empirical coverage is 86.2% for nominal 90% intervals and 91.5% for nominal 95% intervals. Predictive intervals are on average 13.6% wider than the oracle at the 90% level, while CRPS is 14.4% higher than oracle CRPS. Performance is closest to the oracle under autoregressive, threshold, stochastic-volatility, heavy-tailed, and variance-break dynamics, while local trends and regime switching are more challenging. The results indicate that Rollcast can construct competitive probabilistic forecasts from simple, interpretable local summaries, while also identifying limitations in calibration and recursive uncertainty propagation.

Large language models demonstrate increasingly strong reasoning capabilities through effective post-training. Yet, prevailing post-training methods optimize over massive numbers of tokens, implicitly assuming that effective learning must be token-intensive. We revisit this assumption in the on-policy distillation (OPD) setting, which naturally admits dense teacher supervision at every generated token. Using the Qwen3 family, we discover a counter-intuitive phenomenon: reasoning can be effectively incentivized by an extremely small fraction of generated tokens–as few as one or two tokens per reasoning trajectory, corresponding to only 0.05% of all tokens. Surprisingly, this sparse supervision in most cases matches or surpasses full-token training in improving reasoning ability, despite excluding the vast majority of generated tokens from the training objective. This phenomenon is consistently observed across nine teacher–student configurations spanning different model scales on mathematical reasoning tasks, and is further validated on coding reasoning, Llama models and Proximal Policy Optimization (PPO)-based reinforcement learning with verifiable reward (RLVR). Interestingly, such extremely sparse supervision may be closer to the natural learning process: rather than correcting every step word by word, one reflects on a few critical reasoning steps, updates prior understanding, and continues the trial-and-error, avoiding micro-level corrections while remaining remarkably effective. Overall, our results challenge the assumption that effective post-training must be token-intensive and point to a new direction for understanding and designing more efficient post-training algorithms.

Quantum error-correction decoders are typically benchmarked against synthetic circuit-level noise, under the assumption that a decoder's ranking under such noise transfers to hardware and improves as the noise model becomes more realistic. The Willow processor, the first to operate below the surface-code threshold, allows us to test this assumption. We rank a panel of six decoders using a four-rung ladder of noise models with increasing fidelity, evaluated against real data across three code distances, two bases, and fifteen round counts. Rank agreement with hardware appears once the noise model gives each operation type its own error rate. Calibrating the model to the device improves absolute error rates but not rank agreement. We additionally provide the first independent evaluation of NVIDIA's Ising pre-decoder on hardware, at code distances below its training receptive field and via a mapping onto the lattice on which it was trained. Under these conditions, it holds no accuracy-latency advantage: another panel decoder matches or improves on it in both per-cycle error rate and decode latency in 278 of the 280 evaluations. We release the full pipeline and the per-shot outcome of every evaluation, so future decoders and devices can be compared.

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.

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.

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 develop an energy-based reduced-order model for micromagnetic magnetization dynamics that couples a convolutional autoencoder to a structured latent neural ordinary differential equation. Motivated by the precessional-dissipative structure of the Landau-Lifshitz-Gilbert equation, the latent vector field is generated from the gradient of a learned scalar potential through an antisymmetric operator and a symmetric positive-semidefinite dissipative operator. This potential is learned in nonunique latent coordinates and is not identified with the Gibbs free energy, but decreases monotonically along autonomous continuous-time solutions, while the antisymmetric component permits motion along its level sets. The encoder, decoder, latent energy, and operators are trained jointly on short trajectory windows using latent and decoded-rollout losses alone, without time-derivative supervision, physical-energy labels, or dissipation penalties. At inference, an initial state is encoded once, evolved in latent space, and decoded only at the requested output times, enabling substantially cheaper trajectory prediction than the micromagnetic solver used to generate the training data. We compare quadratic, deep, and additive deep-quadratic latent energies on two datasets parameterized by field amplitude and generated for the two applied-field directions of the NIST $\mu$MAG Standard Problem 4. Dissipative-only and antisymmetric-dissipative models achieve comparable accuracy on short training-style windows but differ substantially on uninterrupted rollouts, for which the antisymmetric-dissipative models provide markedly more accurate trajectory predictions. The deep-quadratic energy gives the best overall accuracy for both field directions and exhibits slower error growth when rollouts are extended to twice the training horizon.

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.

We present Hakken, a domain-agnostic prediction and explanation system performing knowledge prediction, i.e., growing scientific knowledge by establishing novel relationships, ones that are not limited to the deductive hull of previous knowledge. Hakken uses a transformer-based prediction model built on temporal sequences of knowledge graphs extracted from vast bodies of research publications, fused with an LLM's semantic knowledge, to predict the presence and define the type of as-yet undocumented relationships between scientific concepts. It then calls a model-agnostic explanation framework to provide accompanying information for each prediction that allows scientists to evaluate the suggested new relationship. While general purpose, we demonstrate Hakken's practical capabilities by applying it to the biomedical domain. There, Hakken's prediction model establishes a new benchmark for time-aware multi-label relation prediction, and we show that the model's output stays coherent and informative over extended time spans in historic data. In addition, we scored 1.5 million above-confidence-threshold hypotheses related to aging, qualitatively validated batches of these predictions with biologists and progressed three of them for empirical validation in wet-lab. Two predictions with potentially significant impact in the context of drug discovery and repurposing were confirmed, introducing previously undocumented interactions between TP53 and BAMBI, and between RAF1 and TNF, to biomedical science.

We present ResLearn-XR, a residual learning framework for predicting eXtended Reality (XR) network traffic and estimating Quality-of-Experience (QoE) risk. ResLearn-XR adopts a two-stage temporal learning structure comprising a base sequence prediction model augmented with task-specific residual learning components to improve adaptability to bursty, non-stationary XR traffic dynamics. The residual learning stages operate in the value space for continuous XR traffic forecasting and in the logit space for probabilistic QoE risk estimation. \rev{For the QoE-risk branch, we introduce a Data Descriptor Algorithm (DDA), a causal feature-construction module that converts packet-level application-layer observables into frame-timing-aware descriptors suitable for encrypted traffic analysis. We also construct an XR Traffic-QoE dataset that pairs continuous XR traffic traces with session-level user-reported QoE labels. ResLearn-XR reduces SMAPE by up to 17.84% across frame-count, frame-size, and inter-arrival-time prediction, while reducing QoE-risk estimation SMAPE by up to 87.8% over single-stage baselines.

Quantization is widely used to reduce the computational and memory demands of neural-network inference. In recurrent networks, however, the quantized state is stored and returned at the next time step, so the rule used to store that state can alter subsequent computations. Here, we introduce recurrent-state write-back to denote this rule and isolate its effect in a compact GRU encoder–decoder for fluorescence lifetime imaging, a molecular imaging modality used in quantitative biological imaging. A central task is estimating two lifetime parameters, the short-lived component {\tau}1 and the long-lived component {\tau}2, from high-noise time-resolved fluorescence signals. Holding the trained model fixed, replacing continuous state propagation with deterministic 4-bit state storage increases estimation errors for {\tau}1 and {\tau}2 by approximately 70x and 300x, respectively. Failure occurs when repeated small updates remain below the write threshold, leaving the stored state nearly fixed while the network continues to propose change. Error feedback, residual memory, and direction memory carry information from these suppressed updates across time and recover accuracy without retraining. Precision sweeps show that increasing state precision can worsen a fixed recurrent solution, while matched training shows that compatibility with the state interface can be learned. To test whether this behavior extends beyond the GRU, we repeat the post-training intervention in an independently trained LSTM, where coarse write-back reproduces the failure, error feedback restores accuracy, and state-specific interventions reveal greater sensitivity of the cell state than the hidden state. Our results establish recurrent-state write-back as a key determinant of low-precision recurrent dynamics and identify the state-storage interface as a central design consideration for quantized recurrent inference.

Federated unlearning aims to remove a client's data from a shared model without retraining from scratch. Some efficient systems make deletion exact by storing compact, additive summaries of the training features and broadcasting an updated linear classifier after every accepted change. We show that these broadcasts can also reveal the hidden summaries. A malicious client can submit known changes, use the returned classifiers to identify the server state, and compare states immediately before and after an isolated deletion. This exposes the deleted sample, class, or client summary and can enable its reinsertion. We characterize exactly when the observations contain enough independent information, give a matching optimal construction for unrestricted probes, and derive a more realistic estimator based on additions formed from the attacker's own data. On MNIST and CIFAR-10, high-precision broadcasts permit exact label recovery for every tested sample deletion with both probe types. Lower-precision broadcasts sharply reduce fine-grained recovery, and insufficiently diverse responses prevent identification altogether. Unrestricted probes are readily detected by their size; most individual attacker-data additions resemble honest batches, although we do not claim that the complete sequence is inconspicuous. The results identify a concrete privacy and integrity risk, its algebraic cause, and practical limits involving broadcast precision, update verification, response rate, and concurrent activity.

In Geometric Deep Learning, inductive biases serve two primary functions: enforcing manifold constraints and embedding relational priors. Currently, representation learning on SPD manifolds frequently relies on pullback Euclidean metrics, such as the Log-Euclidean Metric, to satisfy the former. While computationally efficient in avoiding domain boundary violations, these metrics induce a flat geometry that may fail to capture the intrinsic relational priors of datasets. While metrics such as the Poincaré metric are widely utilized to induce domain-aligned relational priors, generalizing them from standard vector representations to the SPD manifold has remained a challenge. To bridge this gap, we introduce a Nested Inductive Bias framework that utilizes a two-stage diffeomorphic composition to formally pull back non-Euclidean target geometries onto the SPD manifold. This framework enables the construction of curvature-aligned Riemannian classifiers that simultaneously respect matrix constraints and the latent relational geometry of the data. Empirical evaluations on kinematic and signal processing benchmarks, together with synthetic experiments, demonstrate that deep manifold networks experience degradation in class separability unless the metric curvature aligns with the intrinsic data distribution. Furthermore, for standard vectorized architectures, we propose the Rational Conformal Metric (RCM), designed to establish state-of-the-art geometric robustness against outliers by bounding the representation space.

Data collected during aerial and spaceborne imaging spectroscopy campaigns enables the detection of transient events such as trace gas emissions. However, current processing pipelines depend on slow, on-the-ground processing, which delays the time to information of each detected event and prohibits immediate follow-up actions. During the Tokyo Field Campaign of March 2026, we explored on-board processing of Imaging Spectroscopy data from the equipped AVIRIS-5 sensor. Due to communication bottlenecks, full datacubes cannot be downlinked immediately during the flight. Instead we downlink the potential events predicted by our efficient and small machine learning model. We show the first on-board detection of methane point source emission with Imaging Spectroscopy data using Edge ML.

Expert pruning reduces the memory and serving cost of Mixture-of-Experts (MoE) models by removing low-importance experts identified by the router, assuming router probabilities provide a reliable importance signal. We observe that this assumption breaks down under over-dispersed routing, a regime associated with aggressive load-balancing during training, in which tokens are distributed nearly uniformly across experts and importance signals collapse. In this regime, perplexity does not predict downstream task accuracy: on gpt-oss-20B, the lowest-perplexity pruning configuration yields the worst mathematical reasoning, while the highest-perplexity configuration preserves it. This does not occur under standard routing (e.g., Mixtral-8x7B-Instruct), where perplexity and accuracy degrade together. Pruning under over-dispersed routing also exposes a capability trade-off in which no single scoring metric dominates: activation-aware scoring preserves mathematical reasoning but severely degrades knowledge-intensive science (an 18-point gap on GPQA), whereas frequency-based scoring exhibits the reverse. We propose Minimax Expert Score Allocation (MESA), a domain-aware method that iteratively boosts importance scores for experts serving whichever domain is currently worst-affected, minimizing worst-case domain degradation rather than average accuracy. At 25% expert pruning MESA achieves the smallest worst-case degradation across domains, outperforming activation-aware baselines on 7 of 11 benchmarks at a correspondingly reduced memory footprint, and it generalizes to gpt-oss-120B, Gemma-4-26B-A4B, and OLMoE-1B-7B. Our results indicate that over-dispersed routing is a qualitatively distinct pruning regime in which standard assumptions fail, and that recognizing it is a prerequisite for principled expert pruning of load-balanced MoE models.

A conformal certificate can be valid when an LLM answers alone and invalid when the same LLM sees peers that unanimously assert a wrong answer. The question is unchanged; the model's score for the correct answer changes. We call this a score-mechanism shift: clean calibration certifies how the model scores answers alone, but not how it scores them under peer pressure. We show that this shift silently breaks conformal prediction in multi-agent LLM systems. Across open-weight models and multiple-choice QA tasks, coverage falls from a calibrated 90% to 74% under unanimous-wrong peers at the standard alpha = 0.10 operating point. The average hides a sharper failure: by targeting the low-confidence items the certificate still covers, an attacker nearly halves coverage on that subgroup, from 87% to 47%, while the monitored average remains much higher. The failure also reaches the decision layer: a system that should escalate when uncertain can instead become confident enough to act on the attacker's wrong answer. Standard conformal fixes do not solve the problem, because the question distribution has not changed; the model's scoring behavior has.

Transferable coarse-grained (CG) force fields compress chemical space: by aggregating atoms into a reduced set of interaction beads, models such as MARTINI reduce the number of distinguishable compounds by roughly three orders of magnitude, making high-throughput screening of thermodynamic properties tractable across soft matter, with drug–membrane permeability as a well-developed example. The compression is lossy and, so far, one-way: a screen returns a combination of beads, with no established route back to the compounds it stands for. Recovering those compounds–compositional backmapping–is a one-to-many inverse map, distinct from the better-studied conformational problem of rebuilding atomic coordinates from a known mapping. Here we formulate compositional backmapping as conditional graph generation by introducing juniper, a discrete denoising diffusion model over molecular graphs conditioned on the octanol–water partition free energy $\Delta G_{\mathrm{W} \mapsto \mathrm{O}}$, the principal driver of MARTINI bead type assignment and hence a proxy for bead identity. Trained on molecules of up to 9 heavy atoms mapped onto one or two beads, juniper generates molecules that are 93% valid and 92% unique for two-bead targets, and whose $\Delta G_{\mathrm{W} \mapsto \mathrm{O}}$ distributions track the target $\Delta G^{\mathrm{CG}}_{\mathrm{W} \mapsto \mathrm{O}}$ linearly ($r^{2} \geq 0.96$), departing only in the hydrophobic and hydrophilic tails. Although the model receives no chemical information beyond a single scalar, the functional groups shift systematically with the imposed free energy, from branched hydrocarbons at the apolar end to amides, imides, and isocyanates at the polar end. A bead combination flagged by a CG screen can therefore be turned into candidate molecules for atomistic study or synthesis.

Large language models (LLMs) are increasingly used to annotate cultural texts at scales that are impractical for human coders. However, before their outputs are treated as measurements of latent social constructs, it is necessary to establish whether those measurements are reliable. This study evaluates five LLMs as zero-shot annotators of four social constructs expressed in English song lyrics: self-esteem, self-control, seeking belonging, and seeking recognition. Using repeated annotations of a large lyric corpus, we examine three properties of LLM-based measurement: consistency across repeated runs, convergence across models, and transferability of consensus labels to supervised classification. The findings show that LLM-based measurement is not uniformly reliable across constructs. Self-esteem exhibits the strongest repeated-measurement reliability across models, while seeking recognition is generally less stable; self-control and seeking belonging show intermediate but model-dependent reliability. Downstream classification further indicates that consensus LLM labels contain learnable signal, although transferability does not itself establish construct validity. Repeated-measurement stability and cross-model convergence should therefore be reported before LLM annotations are treated as scalable measurements in cultural analytics.

Forecasting-model selection remains difficult in heterogeneous demand because the most suitable decision rule may vary with demand structure, data availability, and forecasting horizon. This study examines whether the selector itself should be treated as a context-dependent component of the forecasting process. Five selection mechanisms - RMSSE, ERA, OWA, CCG-AHSC, and CCG-AHSCD - are compared across 24 optimized forecasting models, nine datasets, three training-testing partitions, and horizons from 1 to 12 cycles. Selector performance is evaluated ex post using Global Relative Accuracy (GRA), statistical tests, and a best-attainable-model reference. No selector dominates across all conditions. CCG-AHSC and CCG-AHSCD are more competitive for Smooth demand and several Erratic configurations, whereas OWA and ERA perform better in Intermittent and Lumpy settings. Selector suitability also changes with historical data availability and horizon, supporting a context-dependent rather than universal approach to forecasting-model selection.

Distributed AI training involves recurring rounds of data exchange between multiple pairs of GPU nodes. Slowdown in even one flow due to congestion can cause the entire communication round to slowdown. Current approaches for evading congestion in AI clusters assume global control over the entire workload (e.g. coordinating the schedule of all jobs) or assume infrastructural support (e.g. adaptive routing in switches). They are thus ill-suited in a shared cloud setting where AI jobs belonging to one user can face external congestion from other users' jobs or background traffic beyond its own control. In this paper, we build a system, REACT, that tunes the recurring pattern of data exchange between GPU nodes (known as communication collectives) in response to congestion. REACT works at the application (communication library) layer, where it detects congestion at runtime using readily available flow stats, and tunes the collective pattern to alleviate congestion - changing the set of incident flows while retaining the semantics of information exchange (e.g. selecting which node aggregates data in an AllReduce tree). REACT requires no explicit support from the underlying network infrastructure and can be unilaterally deployed by individual users in a shared cloud setting. We prototype REACT as a shim layer over NCCL, and evaluate it on a shared academic GPU cluster - enabling REACT improves communication performance (algorithm bandwidth) by 13%-38% under network congestion. Our simulations across a range of congestion scenarios further reveal up to 75% performance improvement, highlighting the effectiveness of our approach.

Learning rich medical concept representations is essential for EHR prediction. Text-attributed knowledge graphs (TKGs) provide a natural foundation by organizing heterogeneous medical relations together with textual semantics. However, most existing encoders process concepts uniformly across patients, despite the fact that a code's meaning and predictive value depend on patient-specific clinical context and trajectory. Learning patient-personalized concept representations from TKGs introduces two key challenges: (1) deciding how much KG context to incorporate for each observed code, and (2) aligning semantic information with the patient-specific relational structure. We propose REFINE, a KG-aware budgeted LLM graph refinement framework for patient-personalized medical concept encoding. Starting from a global TKG, REFINE constructs patient-specific temporal graphs. A sequential reinforcement learning policy selects a personalized KG expansion budget for each observed code. The resulting patient graph is processed by a heterogeneous GNN to capture relation-aware structural dependencies, while a frozen LLM uses graph-aware soft prompts to semantically refine concept representations. Experiments on MIMIC-III and MIMIC-IV show that REFINE consistently improves diverse EHR backbones, outperforms strong baselines, and demonstrates robust gains across component ablation, KG selection, and data insufficiency.

Deep neural operators learn mappings between input functions and complete PDE solution fields, enabling forward evaluations of new problem instances orders of magnitude faster than conventional numerical solvers. Attention mechanisms have recently been introduced into neural operators, but most studies change several architectural components at once, making it difficult to identify what actually improves accuracy. This work presents a controlled and systematic study of five deep operator network (DeepONet) variants with distinct attention mechanisms, trained under both data-driven and physics-informed regimes, to isolate the effects of cross-attention, self-attention, tokenization, and attention depth. We evaluate them on a source-driven transient one-dimensional nonlinear diffusion-reaction equation, a transient one-dimensional viscous Burgers equation with variable initial conditions, and a two-dimensional Poisson heat-conduction problem with heterogeneous source fields. Per-sensor tokenization with cross-attention reduces the mean relative L_2 error of the classical DeepONet in all benchmark-training combinations by factors of 2.4-28.0, while the best attention configurations reach 3.5-32.3. Branch self-attention paired only with dot-product fusion is inconsistent, degrading the one-dimensional problems while helping the more complex two-dimensional source field; added on top of cross-attention it improves all six cases, though by less than cross-attention fusion alone. Global pre-mixing provides no consistent benefit. Increasing cross-attention depth further improves accuracy, but with diminishing returns and a substantially higher cost under physics-informed training. Overall, query-dependent cross-attention is the most reliable mechanism, whereas branch self-attention is most useful for large, spatially complex functional inputs.

Adaptive network intrusion detection systems retrain classifiers after drift alarms, but an alarm detects change; it does not establish that a challenger should replace the deployed incumbent. Promotion is security-relevant because it changes the model responsible for subsequent attack detection, and evaluating it has a methodological problem: promotion conclusions may depend on how the challenger was constructed and on how much evidence supports it. We test that dependence on CICIDS2017, UNSW-NB15 and ToN-IoT with self-contained challenger pipelines, nested candidate-size controls, a common-harness comparison of nine update policies, and a final sensitivity confining every exact feature vector to one evaluation, training or probe role. Incumbent-owned frozen preprocessing amplified apparent promotion harm; with self-contained challenger pipelines the mean full-drift harm did not persist. Raising nominal candidate evidence from 512 to 2,000 samples per class improved promotion under pool-constructed progressive drift by +0.53, +1.67 and +0.38 balanced-accuracy points: positive and statistically resolved in all three benchmarks, but materially benchmark-dependent rather than homogeneous, and driven mainly by fewer false positives. Policy conclusions were partially robust: policy ordering changed with candidate comparability, no policy globally dominated, and earlier compatibility statements for a label-free estimator and a calibrated ensemble narrowed. Validation helped evidence-disadvantaged challengers but added no average benefit at parity. Thirteen replays on real, time-ordered traffic showed no net harm from always deploying. Challenger construction and evidence should be controlled, reported and interpreted explicitly when promotion is evaluated.

We present a systems-security case study of a two-node split-LLM training system whose privacy evaluation passed while leaving an observable channel untested. The Trusted Local Node (TLN) sends protected activations to the Untrusted Cloud Node (UCN), the UCN returns its output, and TLN, holding the private loss, returns the output gradient. The frame the UCN receives mixes real rows with decoys, and the loss ignores the decoys. Their gradients are exactly zero, so the pattern of zeros reveals which rows were real. We measure it with a protocol fixed in advance: a leak injected at known strength to prove the instrument can see one, a shuffled-label control to prove it does not report absent leaks, and a threshold set before the runs. Across nine seeds, the zeros identified the real rows on every frame, 4,096 of 4,096 per run. An attack on the frame contents recovered about one extra token per hundred over a constant-guess baseline (+0.65 to +1.50 percentage points); the shuffled controls recovered nothing. A second set of runs repeated this on a configuration that keeps model quality within budget, so the finding is not confined to a setting nobody would deploy. On both datasets, every such run passed the forward-channel privacy check and the quality check, yet failed that same check once the returned gradient was included. Clipping and noising each row of the gradient closed the leak for about 0.01 nats of held-out cross-entropy. The system is not thereby safe: five classes of attack, including those accumulating observations across training steps, were never measured.

This paper considers the energy landscape of the t-SNE algorithm. While this algorithm has enjoyed broad adoption, the non-convexity of the associated energy has made it difficult to rigorously understand what the algorithm captures in many settings. In particular, a number of well-known numerical examples, several of which are reproduced in this article, suggest a complicated energy landscape with many local minimizers that do not respect the topology or clustering structure of the underlying data. This work seeks to provide first steps towards a rigorous explanation of these phenomena. Specifically, for a general family of energies, which include both the original t-SNE algorithm and recently identified large data limits, and for densities in feature space which obey a continuous symmetry, we construct infinite families of distinct critical points. These critical points are based upon identifying pairs of discrete symmetries, one in the original feature space and the other in the target embedding space, which are preserved under gradient dynamics. These critical configurations exhibit many characteristics, such as topology breaking and spurious clustering, which are often observed empirically. Finally, numerical and analytical examples are given throughout as a means of illustrating the approach.

Pretrained vision-language-action (VLA) models enable broad manipulation but remain unreliable in tasks demanding precision and repeatability. Applying real-world online reinforcement learning (RL) to VLA post-training enables autonomous trial-and-error improvement beyond demonstrations alone, but exposes two bottlenecks: 1) unreliable value signals can induce policy drift; 2) large-VLA overhead constrains throughput and sample efficiency. To address these challenges, we present VLA-Precision, an efficient real-world online RL framework featuring the Asymmetric Co-Bootstrapping (ACoB) algorithm and the ACoB-Stream architecture. Specifically, ACoB establishes asymmetric co-bootstrapping across timescales: early intervention-guided behavioral learning rapidly improves policy performance while enhancing online experience quality. As autonomous experience accumulates, global return propagation and local preference ranking progressively calibrate value estimates, yielding relative action advantages for reference-regularized policy improvement while suppressing drift. To enable ACoB on large VLAs, we develop ACoB-Stream, a closed-loop experience–policy architecture that establishes invariant-state decoupling and on-demand streaming as design principles, delivering up to 10.9$\times$ improvements in throughput and computational efficiency. Extensive evaluations on nine high-precision chemistry tasks across four categories and four robot embodiments show that VLA-Precision achieves 98.3% mean success rate in 45.8 min/task, with 27.6 s episodes running at 1.2$\times$ and 1.8$\times$ the speeds of VLA and RL baselines. Resources are available at https://vla-precision.github.io.

We introduce a lightweight Quantum Prototypical Recurrent Unit (QPRU) that requires significantly fewer parameters than both classical recurrent architectures, such as Long Short- Term Memory (LSTM) and Gated Recurrent Unit (GRU), and quantum variants, including Quantum LSTM (QLSTM) and Quantum GRU (QGRU). Despite its compact design, the QPRU achieves competitive forecasting performance, matching state-of-the-art baselines while offering important structural and practical advantages, including enhanced scalability and a reduced number of trainable parameters.

Large-scale logistics networks require synthetic data generation capabilities to support scenario-based planning under novel conditions-such as network reconfiguration and demand shocks. Existing approaches, which rely primarily on historical observations, lack the ability to generate demand patterns that adapt to changes in network topology while respecting operational constraints. We propose a constraint-aware conditional generative framework for synthetic origin-destination demand generation in hierarchical logistics networks. The framework models demand as a conditional distribution over destinations given each origin, enabling topology-aware synthesis that is both topologically realistic and operationally feasible. Operational guidance is incorporated directly into the generative objective via differentiable constraints, while a flexible conditioning mechanism supports various operational contexts and adaptation to evolving network configurations. We instantiate the proposed framework based on a conditional generative model. Experimental validation on industrial real fulfillment and transportation network demonstrates 16% improvement over graph neural network baselines, 87% operational compliance, and efficient cold-start adaptation, enabling applications in capacity planning, network design evaluation, and routing optimization.