Science & Mathematics

Large language models (LLMs) can improve honeypot realism by generating dynamic, context-aware responses, but powerful cloud models incur substantial cost and latency, while lightweight models may provide weaker deception. This forces practitioners to choose between operational efficiency and interaction fidelity. In this work, we present HydraPoT, a configurable multi-agent honeypot framework that integrates traditional emulation, lightweight LLMs, and powerful cloud LLMs. HydraPoT routes each attacker command to an appropriate agent according to a configurable complexity-based policy, while maintaining shared session state to preserve consistency across agent transitions. We evaluate HydraPoT through an architectural-overhead analysis, a cross-agent context-preservation benchmark, and a three-agent case study. HydraPoT introduces less than 5% runtime overhead, preserves context across heterogeneous agents, and achieves fidelity close to a cloud-LLM-only honeypot. Under our case-study policy, it reduces cloud-processed tokens by 72.6% and operational cost by 3.7x. These results demonstrate that configurable multi-agent routing can make realistic LLM-assisted honeypots more practical under limited operational budgets.

Real-world datasets are inherently class-imbalanced, leading models to favor majority classes and struggle with minority ones. Existing logit adjustment methods address this by modifying logits based on class frequencies, but their class-wise adjustments ignore instance-wise confusion and often suppress informative signals from confusing non-target classes. This results in insufficient learning of such classes, increased prediction uncertainty, and limited generalization. To address this, we propose two complementary methods. First, \textbf{Confusion-based Instance-wise Logit Adjustment (CILA)} adaptively adjusts logits based on instance-wise confusion between the target and non-target classes, restoring informative gradients that class-wise adjustment suppresses. Second, \textbf{Batch-adaptation Noise-Injection (BNI)} perturbs the target logit during training, extending the confusion-aware update to correctly classified but near-boundary samples that the penalty alone cannot reach. CILA is a loss-agnostic plug-in module that can be seamlessly integrated into existing long-tailed learning methods, and consistently improves performance across multiple benchmarks. Our code will be available at https://github.com/maya0395/CILA.

Connected and Autonomous Vehicles (CAVs) face significant challenges in roundabout navigation due to dense traffic interactions and unpredictable agent behaviors, necessitating high levels of adaptability and safety. While the combination of Reinforcement Learning (RL) and Model Predictive Control (MPC) offers a promising approach, existing frameworks suffer from a structural mismatch where pointwise RL decisions fail to align with MPC's trajectory level optimization, and often lack rigorous safety guarantees. This work proposes RLTrajGen-PPMPC-CBF, a hierarchical framework for safe, smooth and comfort roundabout navigation, which regulates the coupling between RL-based planning and MPC-based control. We employ Prescribed Performance Constraints (PPC) to enforce a gradual, stable convergence between reference trajectories and control outputs, ensuring MPC tracking performance remains consistent with intended behavior from RL. Additionally, a Control Barrier Function (CBF) layer provides real-time safety corrections that are fed back into the RL process. Simulation results in the MetaDrive environment demonstrate that our method achieves a 6.25\% increase in average success rate and a 65.87\% reduction in Motion Sickness Dose Value (MSDV) compared to the optimal baseline, meanwhile accelerating training convergence.

Recursive self-improvement, in which frontier AI models work on training the next generation, is quickly becoming a reality. However, the safety of this process has received far less attention than it should. A misaligned model training its successor can shape what that successor knows, how it behaves, and what it wants, and the harm carries forward into the next generation. We study this risk through the lens of AI control. We introduce RSI Control Arena, a testbed in which a frontier model, acting as an autonomous R\&D agent, is given a benign main task from a stage of a successor model's training pipeline together with a covert side task to install a misaligned property in the successor. The testbed contains 16 side tasks in three classes, forming 22 task pairs with main tasks that span the pipeline from data curation to reinforcement learning. We evaluate six frontier models as agents against LLM monitors that read the transcript of the run and the agent's reasoning. We observe that every model completes a substantial share of the side tasks without degrading the main-task performance, including side tasks that make the successor itself a better saboteur. Under minimal instructions to stay covert, models already produce successful attacks that monitors fail to detect, and providing a short list of evasion tactics further reduces the detection rate. Monitors with access to agent's reasoning detects most attacks, but not all. Our results indicate that recursive self-improvement raises significant safety concerns, and that strong safety effort should be in place before models are trusted with greater responsibility of training their successors.

Human-collected real-world robotic manipulation demonstrations often contain mixed-quality trajectories, including hesitant, indirect, or inefficient action segments. While diffusion- and flow-based policies can model such multimodal data, standard behavior cloning (BC) assigns equal likelihood to all demonstrated actions and may therefore reproduce suboptimal modes. Offline RL value estimates offer a natural signal for biasing policies toward higher-value behavioral modes, but directly fine-tuning the generative policy or using critic gradients for policy steering can amplify critic errors and induce prior drift. We propose Advantage-Weighted Latent Steering (AWLS), an offline post-training method for improving frozen one-step generative manipulation policies without updating their decoders. AWLS learns a residual noise shaper, an observation-conditioned residual correction to the policy's latent noise, and trains it with an advantage-weighted action reconstruction objective. The advantages are detached, scale-normalized, and used only as supervised weights on demonstrated actions, avoiding critic-gradient policy updates while preserving fast single-step inference. We further introduce a rollout-free decoder-sensitivity criterion for selecting pretrained checkpoints that remain responsive to latent-noise perturbations. Experiments on simulated and real-world robotic manipulation tasks with clean and mixed-quality demonstrations show that AWLS improves over the frozen base policy while limiting drift from the pretrained behavior prior. Project page: https://awls-corl2026.github.io/

Large language models (LLMs) are increasingly used as judgment assistants and as stand-in participants in psychology experiments, yet little is known about how much their probability estimates move in response to a single new piece of information. We pair premise-hypothesis items from $δ$-SNLI with one strengthener and one weakener each, elicit the prior and the posteriors in separate calls, and measure the directional movement s: the mean posterior change signed by the annotated direction, so that the prior cancels out. We vary whether the update is stated ("Here is something else you learned") or reported ("Someone said that"), and whether the model gives one probability or five weighted estimates. We test seven configurations: two local models (Qwen3-8B, OLMo-2-7B; 240 pairs) and five configurations run through agent command-line harnesses (120 pairs), with hypotheses and analysis rules fixed before the formal runs (not publicly registered). Under stated, direct elicitation all seven move in the annotated direction (s = 18-28 percentage points). Reporting lowers s in all seven point estimates (significantly in six, by 2-5 points) and raises the share of reverse movements in all seven: the effect is consistent but small. Five weighted estimates lower the share of responses at $\leq$5% or $\geq$95% in all seven, but in the two local models they also reduce s by 43% and 68%, whereas in the harnessed configurations the point estimates of the change lie between -3.7 and +0.5 points. An elicitation format that reduces extreme values thus need not preserve sensitivity to evidence; the two should be tested separately. Because $δ$-NLI annotates only direction, we make no claim about how large an update should be. Code, data and the frozen plan: https://github.com/asj252/llm-judgment-studies

Simulating economic responses at society scale requires agents whose decisions reflect household circumstances. We introduce S$^2$-MAS, a society-scale multi-agent simulation framework built on 500 million synthetic household personas grounded in real-survey data. Its Genese engine separates decisions from state updates, while reusable financial scenarios and application ledgers track consumption, cash, debt and payments. Trained with SFT + RLCD on 2.93 million examples, sharp-421m-mlx achieves 74.55% accuracy and 38.37% question-macro F1 on one-shot consumer financial-choice tasks derived from held-out survey responses. Its accuracy exceeds GLM-5.3-Flash, Jev 1.13, DeepSeek-V4.1-Flash, GPT-6-Luna and GPT-6-Astra in the common benchmark (full test for the first four; matched 500-group sample for Astra). We evaluate eight financial task families and simulate 12 million distinct household agents in paired purchase-funding conditions. Jointly raising purchase goals and approved credit limits increases consumption, with responses differing across age and income groups. Education, employment and household-size differences persist after age-income standardization and mainly reflect transitions from deferral to purchase. Asset and debt groups differ in cash drawdown, repayment and closing debt. S$^2$-MAS links heterogeneous household decisions to evolving consumption, cash and repayment paths, supporting economic experiments across household groups and population scales.

Muon and related optimizers update a weight matrix with a spectrally transformed gradient. In this work, we study such spectral optimizers with a constant step size on strongly convex matrix quadratics, which isolate the effect of the update from the confounds of nonconvexity, noise, and step size schedules. We analyze an update using two quantities: the gain, which is the factor by which it amplifies the gradient along each singular direction, and the length, which is the norm of the update. These two quantities place the popular spectral updates in three regimes with different behavior: (i) with a bounded gain, the iterates behave like preconditioned gradient descent and converge for step sizes set by the largest curvature; (ii) with an unbounded gain but a vanishing length, they still converge, with the admissible step sizes set by the total curvature instead; and (iii) with a non-vanishing length, they do not converge for almost every initialization. Interestingly, the practical variants of Muon fall in all three regimes, depending only on how the update is normalized and scaled and not on the polynomial that approximates the polar factor. Experiments validate these findings.

In diffusion models, proper distributional learning entails reversing a diffusion process over underlying density functions, whereas a key failure is memorization, in which the reverse process instead learns diffusion from individual training samples. We provide an analytical characterization of such memorized reverse processes, along with a criterion that detects the associated thermodynamic singularity. The proposed criterion reliably distinguishes between properly learned and memorizing networks across a range of neural network models designed to promote or mitigate memorization.

Efficient use of large-scale supercomputing systems depends on accurate runtime estimation and walltime selection. To avoid premature termination, users often overestimate walltime, leading to poor resource utilization and reduced scheduling efficiency. This paper presents a predictor-agnostic framework for safe walltime recommendation in production HPC environments. Rather than proposing a new runtime prediction model, the primary contribution is the SAFE decision layer, which converts predicted runtimes into conservative bounded walltime recommendations suitable for deployment. Runtime prediction is therefore treated as a modular component within the proposed framework. We evaluate the approach using large-scale production job histories from the Fugaku and NREL Eagle supercomputers, comprising approximately 24M and 11M jobs, respectively. Runtime prediction performance is first evaluated using representative predictors, including linear regression, neural networks, and a neural ensemble. The proposed SAFE framework is then evaluated using the NN ensemble predictor. Experimental results indicate that similarity-aware prediction substantially improves runtime estimation for workloads similar to previous executions. When combined with the proposed conservative walltime recommendation policies, these improvements enable walltime gap reductions of approximately 20 - 26% while maintaining job termination rates below 1% across both systems.

Synthetic aperture radar (SAR) and optical imagery provide complementary structural and appearance information for remote sensing object detection. However, substantial cross-modal discrepancies and indiscriminate channel interaction inevitably introduce redundant or conflicting features. Since different feature channels encode heterogeneous semantic information and contribute unequally to cross-modal fusion, selectively highlighting task-relevant channels while suppressing redundant ones is essential. To address this issue, we propose the Channel Selective Interaction Fusion Network (CSIF-Net), which performs selective channel-level cross-modal interaction. CSIF-Net first employs a frequency decomposition module to construct frequency-aware SAR inputs, facilitating the separation of structural information from noise-related responses. It then models cross-modal channel relationships through a cross-attention channel selection module and dynamically partitions the features into shared and modality-specific components. Finally, a dual-branch fusion module performs cross-modal interaction and channel realignment for shared features while independently enhancing SAR-specific features, thereby exploiting complementary information and suppressing modality interference. Extensive experiments on the OGSOD 1.0 and M4-SAR datasets demonstrate that CSIF-Net achieves state-of-the-art performance in both horizontal and oriented object detection, surpassing the existing methods by 4.01% and 3.18% gains, respectively.

State-of-the-art neural network sparsification relies on alternating dense and sparse training phases. While computationally expensive, dense phases are critical for performance. Our empirical analysis reveals the mechanics behind this gain: upon reintroduction, previously masked parameters receive disproportionately large gradients that drive essential mask rewiring and parameter sign-flipping. Building on these insights, we optimize the performance-to-FLOPs Pareto frontier through three innovations: (1) zero-initializing masked parameters during dense phases to promote relevant sign flips, (2) shortening dense phases, and (3) replacing the dense phase entirely with a sparse rewiring phase that updates only the active mask and a small, high-gradient fraction of out-of-mask weights. Evaluated on CIFAR100 and ImageNet, our variant ARC (Alternating Rewiring Compression) matches baseline AC/DC performance at a fraction of the computational cost, establishing a superior performance-to-FLOPs frontier also compared to standard sparse-to-sparse algorithms. Experiments with GPT-2 on OpenWebText further show improvements in validation perplexity and average zero-shot performance

Latent diffusion models have recently achieved remarkable progress in Real-world Image Super-Resolution (Real-SR). However, existing methods consistently exhibit a persistent seesaw effect: enhancing perceptual realism compromises structural fidelity. This effect, we argue, is not intrinsic to diffusion models, but largely stems from the architectural constraints of prevailing two-stage frameworks. To address this challenge, we propose PixelSR, a novel single-stage Pixel Diffusion Transformer for Real-SR that operates directly in pixel space. Specifically, PixelSR consists of two core designs: Pixel-space Conditional Diffusion and Orthogonal Flow Decoupling. The former establishes an effective pixel-space restoration framework by integrating spatially aligned LR conditioning, visual semantic guidance, and lightweight texture refinement, thereby bypassing VAE compression and enabling end-to-end optimization directly in RGB space. Complementarily, the latter decomposes the asymmetric flow target into a low-rank semantic component and a noise-free orthogonal texture component. By routing these components to specialized branches, it achieves decoupled optimization with reduced mutual interference. Benefiting from these designs, our PixelSR restores perceptually realistic details while faithfully preserving structural content. Extensive experiments across multiple benchmarks demonstrate that PixelSR not only achieves competitive performance, but also effectively mitigates the fidelity--realism seesaw effect.

The UK, EU and US governments have set timelines for post-quantum cryptography migration of critical systems, with deadlines ranging from 2028 to 2035. This paper focuses on IoT-class devices and microcontrollers and the viability of using PQC to connect these devices in secure TLS 1.3 connections. Using wolfSSL, we benchmark a 36-combination TLS 1.3 algorithm matrix, pairing ML-KEM-512/768/1024 and X25519/ECC P-256 key exchange with ML-DSA-44/65/87, Ed25519, ECDSA, and RSA-PSS authentication, across four devices (Raspberry Pi Pico W & Pico 2 W, ESP32, and a Raspberry Pi Zero), measuring both handshake latency and energy consumption. We found that in capable devices, all choices in the TLS matrix gave similar results, but that in IoT-class devices, PQC performance is highly implementation- and architecture-dependent: the common assumption that elliptic-curve key exchange must be cheaper than post-quantum lattice-based key exchange does not hold in this wolfSSL-based implementation, where the best full-PQC combination (ML-KEM-512 + ML-DSA-44) completes TLS handshakes 41% faster than the equivalent classical-modern pairing (X25519 + RSA-PSS) on the Pico W (up to 2.12x as energy-efficient, depending on the paired authentication algorithm), and 8.9x faster on the ESP32. Full-PQC, ML-KEM-512/ML-DSA-44, is not only possible but energy-competitive with or superior to Hybrid or Elliptic Curve alternatives on the most constrained devices, though ML-DSA's variable signing time means applications with strict latency-predictability budgets must accept this variance as the cost of full quantum-safe authentication. We further show that the IETF's draft-specified X25519+ML-KEM-768 hybrid incurs approximately the combined cost of its two components, rather than offering a performance compromise, making pure ML-KEM the more practical choice on hardware lacking Curve25519 assembly support.

Artificial intelligence systems are increasingly deployed in high-stakes decision-making. In response, prototype-based models in explainable AI have attracted growing attention for their ability to retrieve visually similar examples as faithful explanations. Although these models achieve competitive classification accuracy, they often exhibit degraded explanation quality across datasets with varying characteristics, limiting their practical applicability. We draw on insights from cognitive neuroscience, particularly the human memory system, which encodes input-relevant information into representations that support recognition. We propose an Engram Memory Network (EMN), a model that emulates the ventral visual stream and inferior temporal cortex in the human brain. Central to this model is a Hopfield-inspired memory module that iteratively retrieves representative prototypes. These memory-refined latents are used for both reconstruction and classification, encouraging the selected prototypes to capture input-relevant information across diverse datasets. To address the lack of standardized evaluation for prototype explanations, we assess three complementary aspects: perceptual similarity (LPIPS and DISTS) for perceived resemblance, subclass alignment, measuring whether prototypes match the input's fine-grained category beyond supervised labels, and information preservation (NMI), measuring how much input information is captured by selected prototypes. Experiments show that EMN improves visual similarity by 17.06%, subclass alignment by 104.57%, and NMI by 42.42% on average across CIFAR-10/100, MNIST, HAM10000, and CUB-200, while achieving the highest average classification accuracy (87.62%) among all methods. These findings underscore the effectiveness of neuroscience-inspired explanation mechanisms in advancing the interpretability and reliability of prototype-based models.