Searcharxiv⌕ Search

arXiv subjects

N. Benjamin Erichson

Publications and source records attributed to N. Benjamin Erichson.

At least 19 recordsLinked to original sources

The Dose Makes the Poison: Exposure-Aware Evaluation of Generative Drug Design

Generative models are increasingly used to propose molecules for drug discovery. They can produce far more candidates than can be experimentally tested, making computational selection essential. Most selection methods rely on predicted molecular properties, such as potency and toxicity, but overlook how these properties translate into drug exposure at a given dose. As a result, molecules that appear promising on paper may require impractical doses or raise safety concerns. We investigate this gap by using physiologically based pharmacokinetic (PBPK) simulations to re-evaluate molecules selected from three PARP1-targeted generative pipelines. Of 60 shortlisted molecules, seven fail to achieve sufficient target exposure, while 28% of the remaining candidates show predicted safety risks at their required dose. Moreover, 85% of candidates with an achievable dose are outperformed by molecules overlooked during static selection. These findings suggest that accounting for dose and exposure can substantially change which generated molecules are considered promising for drug development.

q-bio.QM↗

Do Reasoning LLMs Refuse What They Infer in Long Contexts?

Long-context LLMs can infer objectives that are not explicitly stated. This capability is useful for reasoning over documents, code, and tool traces, but it also introduces a safety risk. Harmful intent can be distributed across a context and become apparent only after the model combines the relevant information. Existing safety evaluations largely focus on explicitly stated harmful requests and therefore do not fully capture this failure mode. We introduce compositional reasoning attacks, a long-context attack setting in which harmful requests are decomposed into semantically incomplete fragments and embedded in long contexts. The final query is neutral, and the model must retrieve and combine the fragments to infer the implied harmful objective. Empirically, we show that models generally refuse harmful requests when they are explicitly stated. However, refusal rates drop when the same objectives must be reconstructed from fragments, and these failures often become more pronounced in longer contexts. Our analyses suggest that this gap cannot be explained by retrieval difficulties alone. Models can often reconstruct the harmful objective but do not reliably refuse it. Increasing inference-time reasoning effort improves refusal, but at a higher computational cost, with unsafe responses still occurring. Our results reveal a long-context safety gap: Models are better at refusing harmful requests they see than harmful objectives they infer.

cs.CL↗

Advancing Mass-Conserving Deep Learning for Distributed Hydrologic Modeling: A Case Study in a Snow-Dominated Watershed

Recent progress in physics-aware deep learning (DL) has embedded mass conservation within Long Short-Term Memory (LSTM) architecture for hydrologic modeling, improving physical consistency and enabling diagnostics of internal states. However, existing mass-conserving neural networks are primarily developed in lumped formulations. Consequently, previous studies provide only fragmented and often indirect evidence of when and why imposed physical constraints may lead to unrealistic hydrologic behavior. To address this limitation, we propose Physics-Enhanced Conservation-Augmented Net (PECAN), a mass-conserving convolutional LSTM-based framework for distributed hydrologic modeling and diagnosing structural limitations of physics-aware DL models. PECAN is implemented in two variants: (1) PECAN-MC with a strict precipitation-based conservation constraint and (2) PECAN-MCR with a relaxed conservation constraint allowing additional water inputs beyond precipitation. Both variants are trained solely on gauge streamflow and evaluated in a snow-dominated watershed in Colorado. We benchmark PECAN against lumped DL baselines and the distributed process-based Variable Infiltration Capacity (VIC) model. Results show that PECAN achieves competitive outlet streamflow simulation accuracy while producing spatially explicit latent states for interpretation. Comparisons with reference soil moisture indicate that PECAN cell states resemble meaningful temporal soil moisture dynamics. However, analyses also reveal that PECAN misrepresents snow processes by prematurely removing winter snow water storage from the system rather than retaining it for spring release. PECAN demonstrates accurate streamflow simulation at gauges and shows diagnostic potential to expose structural limitations in hydrologic process representation, thereby informing future physics-aware neural network development for hydrology.

cs.LG↗

Coarse-Graining Optimal Transport for Rollout-Free Post-Training of Flow Matching

Flow matching depends critically on the coupling used to pair source and target samples. Optimal transport (OT) couplings can reduce velocity conflict and straighten trajectories; but, in practice, challenges remain: the source comes from a continuous distribution, only a finite number of target samples data are accessible, and the neural network has finite capacity. For these reasons, making the coupling more OT-like may not consistently improve sampling quality, in particular in the low number-of-function-evaluation (NFE) regime. Motivated by this, we introduce ReCouple, a controllable framework that constructs coarse-grained OT couplings, to better understand rollout-free OT-based post-training. We also consider extensions of this basic framework: ReCouple+ extends this toward the continuous-source setting by increasing source resolution, converging to semi-discrete OT in the appropriate limit; and ReCouple++ further adds spatial regularization, motivated by the observation that straight trajectories need not correspond to spatially simple velocity fields. Empirical validation demonstrates the effectiveness of these methods, and it reveals an NFE-dependent interplay between coupling optimality, neural realizability, and generation quality.

stat.ML↗

Generative Refinement Learning for Temporal Interpolation in Scientific Machine Learning

Computationally expensive simulations and real-world observations often provide data at a lower temporal resolution than downstream tasks require. Averaging observations at fixed grid points can introduce errors, and motion estimates alone do not determine how the physical field values change. In this work, we study whether learning to correct a simple numerical interpolation improves the reconstruction of intermediate scientific fields. Our model, interpolative FLow EXpert (iFLEX), first estimates the missing field by taking a weighted average of the two observations at each grid point, giving more weight to the observation closer to the requested time. During training, subtracting this estimate from an observed intermediate field gives the residual that a conditional diffusion model learns to generate. Both observations condition the model, together with the target time, the interval between observations, and available physical parameters. To reconstruct an unobserved intermediate field, the generated residual is added to the weighted average. We evaluate iFLEX on Navier-Stokes Kraichnan turbulence, Shanghai radar precipitation, and sea surface temperature. Across these datasets, iFLEX achieves lower reconstruction error than the evaluated baselines, including optical-flow and diffusion models developed for video frame interpolation in computer vision. In a controlled turbulence comparison at a Reynolds number of 36,000, learning the residual reduces relative Frobenius norm error from about 13.7 for direct field generation to 10.5, with the diffusion backbone and training setup held fixed. Code is available at https://github.com/paulpuren/iflex.

cs.LG↗

Do AI-for-Science Agents Preserve Model Safety?

AI-for-Science agents are increasingly being used for formulating scientific hypotheses, running autonomous experiments, and interpreting evidence. These agents are constructed by integrating a large language model (LLM) into a domain-specific agent harness, which in turn can improve the LLM's scientific capability and enable the model to orchestrate scientific tools. In this work, we ask whether model safety is preserved when an LLM is integrated in an AI-for-Science agent harness. We evaluate five agent harnesses on 50 harmful chemistry requests. Specifically, we study how the safety behavior changes under direct prompting, two simple adversarial prompting strategies, and an adaptive attack. Across four recent GPT variants, our results show that the agent harness can increase harmful compliance even under direct prompting. More critically, we find that adaptive attacks that are largely unsuccessful against the base LLM suddenly become highly successful after the agent harness is introduced. Our results show that the safety behavior of the base LLM is not durable once a harness is introduced. Thus, as AI-for-Science agents become more capable and autonomous, the community needs to evaluate and align the agent using domain-specific safety benchmarks.

cs.AI↗

Multiple Turns, Short Horizons: Measuring Effective-Horizon Compression in Jailbreaks

Multi-turn jailbreaks use feedback from a victim large language model to adapt their queries over several turns. However, once an effective trajectory has been found, it is unclear how much of the earlier query history is still needed. In this work, we study this by removing early attacker queries and running the remaining queries again in a new conversation while regenerating the victim's responses. We define the effective horizon as the shortest remaining query sequence that recovers the same attack progress. Empirically, we find that roughly half of the original query sequence is sufficient on average to recover the same attack progress. We call this phenomenon effective-horizon compression. We show that this shorter history can recover the best point reached during an unsuccessful attack and provide a better starting point for continuing the attack. It also helps identify which earlier queries should receive less learning signal during attacker training, and downweighting those queries improves performance. We further find that short reward-propagation windows preserve calibrated defense performance close to full-trajectory returns. These results show that multi-turn attacks can contain more queries than are needed to recover the progress reached later in the interaction.

cs.AI↗

Is Flow Matching Just Trajectory Replay for Sequential Data?

Flow matching (FM) is increasingly used in scientific domains for time series generation and forecasting, where data often arise from underlying dynamical systems. However, it is not well-understood whether it learns transferable dynamical structure or simply performs an effective ''trajectory replay''. We study this question by deriving the optimal velocity field targeted by the empirical FM objective on sequential data. For the Gaussian conditional paths commonly used in practice, we show that the resulting sampler is an ODE whose dynamics describe a nonparametric, memory-augmented continuous-time dynamical system. The driving velocity field admits a closed-form expression as a weighted mixture of instantaneous velocities determined by observed transitions. This characterization provides an interpretable view of neural FM models as parametric estimation of an underlying nonparametric target and naturally leads to FreeFM: a closed-form, training-free realization of this target. FreeFM provides a transparent reference for studying sequential FM while yielding effective probabilistic forecasts directly from historical transitions.

stat.ML↗

Zatom-2: Multitask Pretraining on Atomistic Data for Generative Modeling across Domains

Unified atomistic modeling has the potential to accelerate discovery in chemistry, materials science, and biology by bridging data-rich chemical domains and data-scarce biological contexts. However, existing generative approaches to atomistic modeling remain highly specialized to scientific disciplines (chemistry vs. biology) or do not leverage both high-volume organic (molecule) and inorganic (material) data for general-purpose pretraining. To this end, we introduce Zatom-2, an atomistic generative model pretrained on approximately five million structures from the OMol25 and OMat24 electronic structure datasets. Zatom-2 features a multiscale Transformer architecture coupled with conditional flow matching that supports force conditioning and foundational pretraining tasks such as generation, structure prediction, and prediction of molecular and material energies and forces. Empirically, Zatom-2 achieves better molecular distribution fidelity than Zatom-1 and achieves strong performance on existing molecule and material generation benchmarks. Zatom-2 demonstrates the ability to control sample generation across low- and high-force regimes, and enhances protein generation in a low-data setting through joint generative-predictive pretraining and transfer learning, increasing protein backbone designability in a length extrapolation setting from 67.8% without pretraining to 74.8% after finetuning on 2,000 protein domains.

cs.LG↗

Variational Streaming Flow: Probabilistic Forecasting in Physical Time

Probabilistic forecasting is important for predicting complex dynamical systems because intrinsic randomness and incomplete observations can cause the same observed state to evolve into multiple plausible futures. While flow matching is a flexible approach for probabilistic forecasting, it is computationally expensive. Streaming flow (SF) reformulates this approach to model temporal evolution efficiently by learning a continuous velocity field directly in physical time. However, SF learns a deterministic velocity field. Thus, it provides only a single future trajectory for a given fixed initial state and observation history. To overcome this limitation, we introduce Variational Streaming Flow (VSF). Our approach learns a latent distribution that is conditioned on the dynamics of interest. In turn, this enables probabilistic forecasting. Importantly, we retain the computational efficiency of SF by generating in physical time. Across deterministic and stochastic dynamical systems, VSF demonstrates superior predictive accuracy and distributional fidelity. We demonstrate the advantage for both long-horizon rollouts exceeding 1,000 steps, and settings with bifurcating dynamics. Moreover, VSF can be integrated into existing Joint-Embedding Predictive Architecture (JEPA)-based world models as a plug-and-play predictor to improve temporal dynamics and goal-directed success rate in navigation, motion planning, and manipulation.

cs.LG↗

LyapuFlow: Controlling Generative Flows with Lyapunov Feedback for Inverse Problems

Pretrained flow models are now widely used as generative priors in science and vision, where inference-time guidance enables test-time constraints without retraining. Existing methods use projection, posterior sampling, or iterative optimization of the generative trajectory. We propose LyapuFlow, an alternative based on Lyapunov feedback control. At each sampling step, LyapuFlow predicts the terminal sample towards which the current flow is evolving, and evaluates the constraint violation on this prediction. Then, we compute the minimum-norm control that satisfies a prescribed Lyapunov decrease condition. The resulting control remains inactive when the uncontrolled dynamics already reduce the constraint violation at the prescribed rate. Otherwise, it provides a corrective update within a feedback trust region that prevents the control from dominating the pretrained dynamics. We demonstrate LyapuFlow in both data and latent spaces, outperforming alternatives spanning different mechanisms for test-time constraint enforcement in scientific machine learning and image inverse problems.

cs.LG↗

AREX: Affine-Residual Exponential Integrator for Few-Step Sampling in Flow Matching

We introduce AREX, a training-free sampler for pretrained flow matching models that uses the target mean and covariance to capture an analytically tractable part of the sampling dynamics. We show that the velocity field of the moment-matched Gaussian target is the $L^2$-optimal affine approximation to the marginal velocity field. This motivates decomposition of the learned dynamics into an affine component over the whole sampling path, determined by the first two target moments, and a neural residual term. AREX keeps the affine component and integrates it using an explicit matrix-valued propagator. In turn, we only require to integrate over the residual term. This differs from scalar exponential integrators, which analytically handle only isotropic linear dynamics. Across image and text-to-image generation tasks, AREX consistently improves sample fidelity in the few-step sampling regime without retraining the underlying model.

stat.ML↗

Divide and Inject: Can Agents Reconstruct an Indirect Prompt Injection from Fragments?

Agentic systems are now being widely used to orchestrate tools and reason over long contexts. However, the improving capabilities of the large language models powering these agents also create new attack surfaces for indirect prompt injection. In particular, an attacker may not need to place a complete malicious instruction in retrieved content if the agent can reconstruct the objective from incomplete fragments distributed across a long context. In this work, we introduce adaptive long-context prompt injection (AdaLCPI), which combines long-context fragmentation with adaptive search. AdaLCPI splits an attack objective into incomplete fragments, embeds them in external content retrieved through the agent's tools, and uses a reconstruction cue to prompt the agent to combine them. It then iteratively refines the fragments and cue with OpenEvolve using graded scoring and natural-language execution feedback from the target agent. Empirically, AdaLCPI achieves higher attack success than strong adaptive baselines, reaching 61.4\% macro-average ASR compared with 32.8\% for Trojan Hippo-style and 30.0\% for AgentVigil. Safety evaluations should therefore test whether agents remain robust when harmful objectives must be reconstructed from incomplete fragments.

cs.AI↗

D-Judge: Disrupting Multi-Turn Jailbreaks using Semantics-Preserving Output Rewriting

Multi-turn jailbreak attacks pose a growing threat to large language model (LLM) safety because they exploit feedback from auxiliary judge models to iteratively refine prompts toward harmful goals. Existing defenses largely detect or block unsafe content at individual turns or at the final response, leaving the judge-driven refinement loop intact and allowing attackers to extract informative feedback from intermediate interactions. We introduce D-Judge, a semantics-preserving output rewriting defense that intervenes directly in this loop by rewriting the victim LLM's responses before they are evaluated by the attacker's judge. By misaligning the judge's feedback signal without changing the meaning of the original response, D-Judge derails the attacker's prompt-refinement process, causing subsequent queries to be optimized against a distorted signal of attack progress. To improve D-Judge's ability to produce such rewrites, we construct a dataset of semantically equivalent response pairs that induce different judge-assigned harmfulness scores, and use it for supervised fine-tuning followed by direct preference optimization. Experiments on HarmBench show that D-Judge reduces the success rate of state-of-the-art multi-turn jailbreaks while preserving performance on benign benchmarks.

cs.CR↗

Reducing the Safety Tax in LLM Safety Alignment with On-Policy Self-Distillation

Safety alignment often improves robustness to harmful queries at the cost of reasoning ability, a tradeoff known as the safety tax. A common cause is distributional mismatch: supervised fine-tuning trains the target model on safety demonstrations produced by humans, external models, or fixed self-generated traces, rather than on trajectories sampled from its own policy. We identify off-policy training mismatch as a second source of this tax and study on-policy self-distillation for safety alignment, which we call OPSA. The model generates its own rollouts and receives dense per-token KL supervision from a frozen teacher copy of itself conditioned on a privileged safety context. Because this teacher must be safer than the sampled student trajectory, we introduce \emph{teacher flip rate}: a criterion that measures how often a privileged context converts unsafe responses into safe ones. We use this signal to search for contexts that activate latent safety reasoning rather than merely elicit safe-looking demonstrations. Across two reasoning-model families and five model scales, OPSA achieves a stronger safety--reasoning tradeoff than off-policy self-distillation and external-teacher distillation under matched data and full-parameter fine-tuning, with the largest gains on smaller models (+8.85 points on R1-Distill-1.5B and +5.49 points on Qwen3-0.6B). The gains persist across training-set sizes and adaptive jailbreak evaluations. Token-level analyses further show that OPSA concentrates updates near early compliance-decision tokens, providing a mechanism for improving safety while preserving general reasoning.

cs.LG↗

Zatom-1: Towards a Multimodal Foundation Model for 3D Molecules and Materials

General-purpose 3D modeling in chemistry encompasses molecules and materials, requiring both generative and predictive capabilities. However, most existing AI approaches are optimized for a single domain (molecules or materials) and a single task (generation or prediction), which limits representation sharing and transfer. We introduce Zatom-1, a cross-domain, general-purpose model architecture that unifies generative and predictive learning of 3D molecules and materials. Zatom-1 is a deliberately simplified Transformer trained with a multimodal flow matching objective that jointly models discrete atom types and continuous 3D geometries. This approach supports scalable pretraining with predictable gains as model capacity increases, while enabling fast and stable sampling. We use cross-domain generative pretraining as a universal initialization for downstream multi-task prediction of properties, energies, and forces. Empirically, Zatom-1 outperforms or competes with specialized baselines on both multi-task generative and predictive benchmarks in data-controlled settings, while improving generative inference speed by more than an order of magnitude. Our experiments demonstrate positive predictive transfer between data domains from joint generative pretraining: modeling materials during generative pretraining improves molecular property prediction accuracy. Open-source code and model weights are freely available at https://github.com/Zatom-AI/zatom.

cs.LG↗

Sharpen Your Flow: Sharpness-Aware Sampling for Flow Matching

Flow matching models generate samples by numerically integrating a learned velocity field, with each integration step requiring a neural network evaluation. Fast generation therefore requires using a small fixed evaluation budget effectively: the key question is not only how to integrate the flow, but where the sampler should spend its steps. We propose SharpEuler, a training-free sampler that profiles a pretrained model offline by estimating where the learned velocity field changes most rapidly along calibration trajectories. This finite-difference estimate defines a solver-aware sharpness profile, which is smoothed and converted by a quantile transform into a timestep grid for any desired inference budget. At test time, sampling remains ordinary Euler integration with the same number of model evaluations as a uniform schedule. We justify SharpEuler using three principles: a numerical principle identifying trajectory acceleration as the leading source of Euler discretization error, a variational principle deriving sharpness-based power-law timestep densities, and a statistical guarantee showing that the finite-sample calibrated sampler is stable at the terminal distribution level. Our experiments show that SharpEuler improves sample quality at fixed budgets, reducing inter-mode leakage and increasing mode coverage.

cs.LG↗

MT-JailBench: A Modular Benchmark for Understanding Multi-Turn Jailbreak Attacks

Multi-turn jailbreaks exploit the ability of large language models to accumulate and act on conversational context. Instead of stating a harmful request directly, an attacker can gradually steer the conversation toward an unsafe answer. Recent methods demonstrate this risk, but they are usually evaluated as black-box pipelines with different budgets, judges, retry rules, and strategy generation procedures. As a result, it is often unclear whether reported gains reflect stronger attack mechanisms or different experimental conditions. We introduce MT-JailBench, a modular evaluation framework for benchmarking multi-turn jailbreaks under fixed conditions. MT-JailBench implements each attack as five interacting modules: evaluation function, attack strategy, prompt generation, prompt refinement, and flow control. This design enables fair comparison across attack methods and component-wise analysis of what drives attack success. Using MT-JailBench, we find that resource budgets and evaluation functions are major confounders: controlling turns, retries, interactions, sampled strategies, and judges substantially change the ranking of attacks. At the component level, prompt generation accounts for most performance variation, while refinement and flow control provide moderate gains. We also find that explicit dynamic strategy generation is not always necessary; stochastic sampling from a fixed strategy can rival more elaborate diversification mechanisms. Finally, recomposing the best components yields a strong attack configuration that outperforms its source attacks and generalizes across diverse target LLMs. MT-JailBench therefore provides a modular framework for comparing multi-turn jailbreaks, understanding the impact of components, and guiding stronger red-teaming evaluations.

cs.CR↗