Science & Mathematics

As people use large language models (LLMs) over time, they develop their own ways of prompting to achieve desired results. These useful experiences often remain scattered and unorganized across conversations. Yet existing personalization approaches primarily capture user preferences or explicitly defined workflows. Through a formative study, we found that users develop prompting strategies that range from implicit instructions to explicit templates from these experiences. We further find that interaction behaviors provide evidence for these strategies, which can be decomposed and recombined for reuse across tasks. Building on these findings, we present Promptile, a framework that identifies and validates users' prompting strategies from interaction histories. Promptile organizes these strategies into reusable components that can be retrieved, combined, and adapted to users' current tasks. Our evaluation shows that Promptile accurately recovers and reuses these strategies, better aligning model outputs with users' expectations while reducing repetitive prompting effort.

Scene-adaptive immersive audio rendering requires rapid estimation of acoustic parameters that reflect the surrounding environment. For repeated runtime scene updates, performing a detailed geometric-acoustic simulation for every update can be computationally costly. We present Point2Acoustic, a learning-based framework for direct estimation of six-band reverberation time ($RT_{60}$) and reverberant-to-direct ratio (RDR) from 3D scene geometry and acoustic material properties. Each scene is represented by a surface-sampled point cloud with spatial coordinates and frequency-dependent absorption and scattering coefficients, and encoded using PointNet or PointNet++. We train and evaluate the framework on a synthetic furnished-room dataset with varied geometry and acoustic materials, using geometric-acoustic simulation to obtain reference targets. Point2Acoustic achieves accurate multi-band prediction with millisecond-scale inference, and ablation results indicate that explicit global room descriptors and semantic surface labels provide no consistent benefit. The proposed formulation supports efficient scene-adaptive acoustic rendering in extended reality (XR).

Generative models for precipitation nowcasting are trained and evaluated predominantly on wet-biased datasets, which over-represent precipitation events to emphasize rare extremes. Selecting test cases based on target observations is known to favor forecasts that match the selected distribution rather than the intended deployment distribution. We show analytically how this target-dependent sampling also shifts the conditional distribution learned by generative models. When training and test sampling match, even an ideal model learns precisely the distribution favored by evaluation, allowing the evaluation bias to conceal the training bias. Using a fixed satellite-to-radar diffusion model, we demonstrate that target-dependent test sampling can reverse the calibration assessment. The model appears under-predictive on a wet-biased test set but over-predictive on a deployment-representative test set. This result is replicated outside the model's training domain after a seven-year gap. Utilizing classifier-free guidance as an example, we demonstrate how the contradictory calibration assessment can misdirect downstream model development.

The limited onboard computing resources of robots make the inference demands of vision-language-action (VLA) models costly to accommodate. Reducing these demands through aggressive compression, however, can degrade manipulation performance beyond what supervised fine-tuning alone recovers. We present SPAR (Structural Pruning and Action Recovery), which simultaneously prunes the hidden and feed-forward intermediate dimensions of the large language model (LLM) backbone through pruning-aware tuning, then applies reinforcement learning (RL) to recover policy performance. Applied to $π_{0.5}$, SPAR reduces the language backbone from 2.34B to 0.09B parameters and the full model from 3.37B to 1.12B, removing approximately 96% and 67% of their parameters, respectively. At this compression level, RL raises average success across the four LIBERO suites from 84.55% to 93.45%, an 8.90-percentage-point gain. The compressed policy achieves a $1.98\times$ full-policy inference speedup over the unpruned model. These results identify substantial intermediate-dimension redundancy and show that post-pruning RL can recover manipulation performance in compact policies.

Machine-learning (ML) malware detectors are known to be vulnerable to adversarial samples. Generative adversarial networks (GANs) such as MalGAN, Mal-LSGAN, and our previous Mal-D2GAN can craft adversarial malware to probe and harden these detectors, yet they operate directly on a high-dimensional, discrete binary feature space, which leads to unstable training, vanishing gradients, and adversarial samples of limited diversity. To address these problems we propose Mal-AAEGAN. First, an adversarial autoencoder (AAE) learns a low-dimensional, regularized and continuous latent representation of the $M$-dimensional binary API feature vector, and the generator performs an additive perturbation in this latent space rather than in the raw binary space. Second, training follows the Wasserstein GAN with gradient penalty (WGAN-GP) scheme using two critics: a mimic-black-box critic that inherits the role of the substitute detector and a naive critic that inherits the role of the additional detector. A decode-binarize-OR pipeline guarantees that the original API calls of the malware are preserved. The model is evaluated on two malware datasets of different input dimensionality ($M\in\{139,215\}$) against eight common ML detectors and compared, under an identical protocol, with MalGAN, Mal-LSGAN and Mal-D2GAN. Mal-AAEGAN trains stably in the continuous latent space and attains an evasion rate on par with or higher than these binary-space baselines across both datasets, while its generator stays responsive to the noise input rather than reducing to a single fixed perturbation per malware.

Changing sparse autoencoder (SAE) codes can alter both intervention direction and magnitude, so comparing intervention effects requires a strength control specifying what is held equal. Norm matching and output calibration provide different controls, but whether interventions yield different effects across codes after strength matching remains unclear. We compare intervention effects under these controls for fixed feature groups in Gemma and Llama and observe: (1) Whole-patch scaling. With one coefficient per input, full-output Kullback-Leibler (KL) calibration leaves smaller mean effect gaps from the encoder-based reference intervention than total-norm matching for nonnegative least-squares (NNLS) refits. On new Llama inputs, total-norm matching loses 11 reference transfers, whereas full-output calibration preserves every transfer decision. (2) Positionwise scaling. On a small matched gender-agreement subset, positionwise norm matching leaves smaller observed mean gaps than full-output calibration. (3) Answer exclusion. Compared with unscaled NNLS, answer-excluded calibration reduces gaps in the measured subject-verb agreement comparisons but increases them for gender refits. Full-output KL calibration also reduces larger gaps for norm-matched constructed candidates, although matching coverage is incomplete in several settings. Since calibration uses the evaluated input's outputs, it serves as a diagnostic of residual disagreement.

In this work, we address real-time fine-grained human activity recognition (HAR) in industrial assembly and introduce PDNet, a new framework that jointly exploits structured pruning and knowledge distillation for 3D convolutional networks. We start by training a 3D-CNN backbone in a supervised manner. Then, PDNet applies channel-wise structured pruning to obtain a more compact network. In the third step, our method is optimized using knowledge distillation from the large unpruned teacher to recover accuracy lost due to pruning. On the challenging MECCANO dataset with an X3D-M backbone, the pruned model without distillation reaches a Top-1 accuracy of 25.54 %, while PDNet raises this to 39.46 %, with only 6.90 MB model size and 3.78 GFLOPs. These results show that distillation within PDNet is highly effective in restoring pruned model performance and highlight its high potential for resource-constrained HAR in industrial assembly.

Reinforcement learning for language models has moved from preference reward models to verifiable rewards, which sustain much longer optimization. Preference models are prone to reward hacking: the policy raises their score without improving response quality, or even while degrading it. Yet for open-ended tasks, where quality can neither be verified nor fully written down as rubrics, they remain valuable. We introduce Dynalize, which mitigates reward hacking online, during policy training. Periodically and asynchronously, a language model analyzes training rollouts and writes penalties for undesirable behaviors that the reward model appears to reward; a judge applies them to every rollout. In clinical counseling and creative writing, training on the reward model alone raises its score continuously but ultimately leads to a decline in external benchmarks; Dynalize sustains longer training and results in improved downstream performance. We find that our method suppresses patterns incentivized by preference models that, without penalties, spread to most responses, and even to the model's reasoning, which the reward model does not explicitly score. Furthermore, we show that its gains hold in multi-turn training, and even after we retrain the reward model on the discovered behaviors.

We study capacitated line aggregation with per-batch maximum delay. Requests arrive over time at locations on a rooted half-line and are served in instantaneous batches of at most $k$ requests. Each batch pays the distance to its farthest request plus the waiting time of its oldest request. The objective is to minimize the total cost. We give a deterministic online algorithm with competitive ratio $36+12\sqrt3$. The algorithm uses a Next Fit idea along paths of nested time intervals: it assigns a new group to an existing batch when a path-load test permits, and opens a new batch otherwise. The path load controls both capacity and the cost of opening new batches. Randomizing the common scale improves the expected competitive ratio to $(18+6\sqrt3)/\ln2$ against an oblivious adversary. Both guarantees extend to arbitrary indivisible request sizes. We also study capacitated LAD in the learning-augmented model. Advice revealed at its service times allows $(1+ε)$-consistency against every complete feasible advice schedule and $O(ε^{-2})$-robustness. The analysis charges repeated switches to successive decreases in the remaining advice cost.

We develop a vision--radio-frequency (RF) cooperative integrated sensing and communication (ISAC) framework that jointly uses visual priors and RF observations for transmit-side beam control and receiver-side clutter suppression. Within a narrowband snapshot model, visual target priors and historical RF detections guide control in dynamic vehicle-to-infrastructure networks with roadside clutter. In this framework, posterior angular uncertainty determines beam control and the selection between communication-for-sensing and sensing-for-sensing modes. To reduce sensing overhead while accommodating unreliable visual information, we develop a reliability-adaptive beam scanning scheme that allocates a bounded scan budget according to visual target probabilities and RF history, while reserving an adaptive fraction for exploration. At the receiver, visual geometry identifies clutter support, and RF observations estimate the corresponding complex reflection coefficients. Motion-aware residual filtering is then applied before angle--Doppler detection to suppress clutter while preserving slow-moving targets. Furthermore, communication and sensing power allocation is formulated as a per-direction linear program, subject to signal-to-interference-plus-noise ratio constraints evaluated at angles sampled from the posterior distribution and a total-power budget. Simulation results show that the proposed framework improves tracking over static and vision-only beam control and reduces sensing overhead relative to full scanning. The proposed scanning scheme also provides greater robustness to visual blockage than fixed-weight scanning, while the receiver better preserves slow-moving targets than temporal cancellation when visual target information is available.

AI agents in containers act by opening network connections, and the text that drives them can be hostile. Controls placed between the agent and its tools are bypassed by a child process or a direct socket, and their approval is not tied to the connection the kernel later opens. We present connection escrow: every outbound TCP connection from a supervised pod is parked inside the kernel's connect() path, before any packet exists, by a 98-instruction eBPF program on the pod's cgroup. A node interceptor attributes the held socket to its kernel socket cookie, process, executable digest, cgroup and pod identity, parks the first bytes the agent writes, and asks a verifier in a separate container with its own key. The connection is released only on a signed verdict naming that one socket cookie and destination, unexpired and unused; no verdict by the deadline resets it, and a late approval is refused. The agent is unmodified. A 36-check adversarial suite with real cgroups passes on the eBPF hold point (33 of the checks also apply, and pass, on an iptables hold point): refused connections never reach the destination; forged, rebound, replayed, expired and late verdicts are refused; the agent's attempt to reach the verifier is itself escrowed; child-process and renamed-binary bypasses fail; UDP is pinned to the resolver. On a Kubernetes cluster, unmodified pods are held per connection and attributed to their SPIFFE identity, a raw socket that bypasses a sidecar proxy is held, and 600 connections from six pods are gated. The hold adds 4.5 ms median per released connection, of which 24 us is policy. Finally, we gate on request content without terminating TLS: an eBPF uprobe reads the plaintext at the agent's TLS-library boundary (shared OpenSSL, static OpenSSL, Go crypto/tls), a model scores it, and the connection is held on that verdict while TLS stays end-to-end and the enforcer holds no key.

Plasma-facing components (PFCs) in future fusion reactors must withstand extreme environments, making tungsten (W) a prime candidate material due to its high thermal conductivity, high melting temperature, low sputtering yield, low tritium absorption. However, neutron irradiation introduces complex nanostructural defects that severely degrade these critical properties. In this work, we use Non-Equilibrium Molecular Dynamics (NEMD) simulations coupled with the Creation Relaxation Algorithm (CRA) to investigate the lattice thermal conductivity (kappa) of W across a wide range of irradiation doses. Systems treated with different doses up to 1.8 canonical displacements per atom (cdpa) were simulated to generate and study diverse defect structures, including vacancies and self-interstitial atom (SIA) clusters. By correlating spatially-resolved temperature gradients with local defect concentrations, we show that even small changes in defect type and distribution significantly affect thermal transport. We employ Machine Learning (ML) models to develop a quantitative relationship between the defect features and the resulting thermal conductivity degradation. The ML analysis identifies vacancy density as the dominant structural factor controlling thermal transport. We found a significant, nonlinear suppression of kappa with increasing defect density. Predictive multi-defect mapping built upon our ML model resolved a step-like degradation transition at a critical vacancy threshold of 0.05 at.%, beyond which phonon scattering completely saturated and capped the maximum relative lattice thermal conductivity degradation at approximately 30 at.%. This integrated simulation and ML approach provides a powerful, predictive framework for quantifying the complex relationship between structure and thermal conductivity in irradiated fusion materials.

Latent memory is gaining ground as a lightweight history supplement for GUI agents. It supplies history to a frozen vision-language model via prefix vectors trained with step-wise cross-entropy (CE). While resulting gains are naturally attributed to encoded history, our controlled experiments show they primarily stem from action revision. The prefix learns to alter the model's default actions. However, this revision is unscoped because step-wise CE pushes outputs toward reference actions at every step without explicitly preserving correct decisions. Consequently, it repairs errors but also breaks 8-13% of initially correct answers. Overall accuracy metrics hide these losses. We therefore propose ScopeMem, a latent-memory method for scoped revision. Its encoder conditions prefix generation on the model's initial answer and is trained to correct errors while preserving correct decisions. On the controlled ablation set, ScopeMem reduces the share of broken correct decisions to 5.6% while repairing more errors than any evaluated CE-trained prefix. On AndroidControl, GUI-Odyssey, and zero-shot AITW, ScopeMem outperforms all evaluated prompting and latent-memory baselines, enabling a frozen Qwen3-VL-8B to match or surpass GUI-specialized agents and Qwen3-VL-32B.

Deployed vision models auto-accept confident predictions, so high-confidence errors propagate silently. Maximum Softmax Probability (MSP), the standard failure predictor, has a structural blind spot: as a scalar read from the classifier head, it is first-order blind to feature variation orthogonal to its confidence gradient. Through a factorial analysis of feature-space detectors, we identify a conditional-redundancy principle explaining why they fail to complement MSP: the class-separability geometry they measure is already encoded in MSP and collapses under fusion. The signal that survives is class-conditional typicality on the off-manifold residual---and because this residual's distribution varies by backbone, from near-Gaussian to heavy-tailed, it must be estimated non-parametrically. We instantiate these findings as CANARY, a training-free detector that computes a per-class kNN distance on the PCA residual of the penultimate feature, percentile-calibrates it on validation data, and fuses it with MSP through a single validation-tuned weight. CANARY's benefit follows a headroom law: gains grow as baseline MSP quality falls, so a single clean-validation number forecasts the improvement. With every parameter selected on in-distribution data alone, CANARY matches or improves 382 of 385 shift cells (e-AURC; 379 on AUROC) across five backbones, with worst-case regression within seed noise ($-0.0002$ AUROC) and the largest gains on natural distribution shifts (ImageNet-A/R, ObjectNet)---where deployed models are most exposed.

A bimanual demonstration specifies both what the robot should do and one particular schedule for when each arm should act. Standard imitation learning does not distinguish task-required cross-arm constraints from incidental scheduling choices, and may therefore exploit opposite-arm progress as a shortcut cue, creating spurious cross-arm coupling even where relative timing is irrelevant for the current task phase. We introduce Constraint-preserving Temporal Recombination (CTR), a training-time data intervention that preserves recorded arm-local sensorimotor content while retiming the corresponding segments to realize admissible schedules that preserve task-required cross-arm constraints. We use CTR to train Diffusion Policy, Action Chunking with Transformers, and $π_{0.5}$, and evaluate them on four simulated and four real-robot tasks. Across six real-robot bimanual progress-state initializations, CTR raises worst-state success from 56% for the strongest non-CTR schedule mixture to 76%. With $π_{0.5}$ on an Independent task, a 4 s pause transfers 2.56 s of delay to the opposite arm under the Concurrent schedule distribution, but only 0.25 s with CTR. On real robots, CTR violates task constraints in only 0--4% of trials, versus 36--98% for narrow-schedule baselines. Code will be released at https://github.com/MINT-SJTU/CTR.