Tracing the tissue and cell-type origins of extracellular vesicles (EVs) in blood is critical for liquid biopsy and precision medicine, yet existing deconvolution methods remain limited by the need for labor-intensive reference signatures and poor adaptability to distribution shifts between tissue/cell-type datasets and EV transcriptomes. We introduce DADA-EV (Domain-Adaptive Diffusion Autoencoder for EVs), a hybrid deep learning framework that combines an autoencoder backbone with a generative simulation module and adversarial domain adaptation. DADA-EV features three key innovations: (1) a reference-free design that eliminates reliance on predefined signatures; (2) cross-domain generalization by aligning feature distributions between source (tissue/cell-type) and target (EV) domain; and (3) reduced dependence on source data during target-domain training. Extensive evaluations on pseudo-EV data show that DADA-EV consistently outperforms existing approaches, yielding accurate fraction estimates across diverse tissues and gene sets. Validation using in vitro cell-line mixtures further confirms its reliability in resolving complex compositions, demonstrating high sensitivity in detecting low-abundance targets. Applied to real EV transcriptomes, it reveals tissue- and cell-type heterogeneity across patient groups. In summary, DADA-EV provides a robust, reference-free, and generalizable solution for EV origin tracing, with strong potential to advance diagnosis, prognosis, and treatment monitoring via liquid biopsy.

Decision-making processes involving fixed charges arise in various real-world applications and can often be modeled as mixed-integer nonlinear programs (MINLPs) with semicontinuous variables. Perspective reformulation, a technique leveraging perspective functions, offers tight formulations for such MINLPs. In this article, we address the challenge of solving such reformulations by introducing perspective Benders cuts, a family of generalized Benders optimality cuts, and compare them with the classic generalized Benders cuts and the perspective cuts. We focus on their applications to two fixed-charge nonlinear resource allocation problems: a generalized sensor placement problem and a generalized uncapacitated facility location problem. The original quadratic allocation cost functions in these problems are extended to a class of reducible convex functions. By leveraging the reducible property of nonlinear resource allocation problems, we develop an ad-hoc procedure of solving the reduced quadratic subproblems to efficiently separate perspective Benders cuts. These features contribute to a highly efficient branch-and-Benders-cut approach, as demonstrated through extensive computational experiments on various sets of benchmark instances. History: Accepted by Antonio Frangioni, Area Editor for Design & Analysis of Algorithms–Continuous. Funding: K. Yang acknowledges financial support from China Scholarship Council [Grant 202406110031]. This work was also supported by the National Science Fund for Outstanding Young Scholars [Grant 62122093], the National Natural Science Foundation of China [Grants 72101264, 72431011, and 72421002], the Science and Technology Innovation Program of Hunan Province [Grant 2023RC3008], Open Project of Xiangjiang Laboratory [Grant 22XJ02003], and the University Fundamental Research Fund [Grant 23-ZZCX-JDZ-28]. H. Yang’s work is funded by National Natural Science Foundation of China [Grant 72201232 and 72231008], Guangdong Provincial Key Laboratory of Mathematical Foundations for Artificial Intelligence [Grant 2023B1212010001], and Shenzhen Key Laboratory of Crowd Intelligence Empowered Low-Carbon Energy Network [Grant ZDSYS20220606100601002]. Supplemental Material: The software that supports the findings of this study is available within the paper and its Supplemental Information ( https://pubsonline.informs.org/doi/suppl/10.1287/ijoc.2024.0984 ) as well as from the IJOC GitHub software repository ( https://github.com/INFORMSJoC/2024.0984 ). The complete IJOC Software and Data Repository is available at https://informsjoc.github.io/ .

Mixed redundancy, which combines active and cold-standby redundancy, can improve reliability design flexibility but substantially increases computational complexity. Consequently, it is rarely used in multi-objective reliability-redundancy allocation problems (MRRAPs), as balancing conflicting objectives proves challenging. To address it, this paper formulates a bi-objective RRAP (BRRAP) with mixed redundancy, aiming to maximize system reliability while minimizing cost. The reliability of cold-standby and mixed redundant subsystems is precisely evaluated using continuous time Markov chain models. Although swarm intelligence algorithms are widely used for MRRAPs because of their global search capability and implementation simplicity, their stochastic updating mechanism often leads to weak local exploitation and premature convergence. To overcome this limitation, an importance measure (IM)-based two-stage BRRAP optimization framework is developed, which iteratively combines swarm intelligence-based global search with IM-guided local refinement. By adjusting Pareto solutions from both reliability and redundancy perspectives, the IM-based local optimization effectively pushes the Pareto front toward higher reliability and lower cost. Experiments based on four benchmarks demonstrate that the proposed framework improves solution quality, convergence, and diversity of Pareto fronts.

Language Model (LM) agents for cybersecurity that are capable of autonomously identifying vulnerabilities and executing exploits have potential to cause real-world impact. Policymakers, model providers, and researchers in the AI and cybersecurity communities are interested in quantifying the capabilities of such agents to help mitigate cyberrisk and investigate opportunities for penetration testing. Toward that end, we introduce Cybench, a framework for specifying cybersecurity tasks and evaluating agents on those tasks. We include 40 professional-level Capture the Flag (CTF) tasks from 4 distinct CTF competitions, chosen to be recent, meaningful, and spanning a wide range of difficulties. Each task includes its own description, starter files, and is initialized in an environment where an agent can execute commands and observe outputs. Since many tasks are beyond the capabilities of existing LM agents, we introduce subtasks for each task, which break down a task into intermediary steps for a more detailed evaluation. To evaluate agent capabilities, we construct a cybersecurity agent and evaluate 8 models: GPT-4o, OpenAI o1-preview, Claude 3 Opus, Claude 3.5 Sonnet, Mixtral 8x22b Instruct, Gemini 1.5 Pro, Llama 3 70B Chat, and Llama 3.1 405B Instruct. For the top performing models (GPT-4o and Claude 3.5 Sonnet), we further investigate performance across 4 agent scaffolds (structured bash, action-only, pseudoterminal, and web search). Without subtask guidance, agents leveraging Claude 3.5 Sonnet, GPT-4o, OpenAI o1-preview, and Claude 3 Opus successfully solved complete tasks that took human teams up to 11 minutes to solve. In comparison, the most difficult task took human teams 24 hours and 54 minutes to solve. Anonymized code and data are available at https://drive.google.com/file/d/1kp3H0pw1WMAH-Qyyn9WA0ZKmEa7Cr4D4 and https://drive.google.com/file/d/1BcTQ02BBR0m5LYTiK-tQmIK17_TxijIy.

Research has characterized mind-wandering as humans’ natural mental state, with moments of task-focused attention being the exception. With this framing, mind-wandering while driving likely occurs more than generally acknowledged, and seems poised to increase with higher levels of automation. This in turn may have adverse effects on drivers’ abilities to regain situation awareness or resume control when needed. Of the prior work on detecting mind-wandering while driving, none focuses on automation or complex urban environments. We ran an exploratory study (N = 14) of an automated drive through New York City in a two-dimensional virtual reality context, focusing on physiological measures such as gaze distribution, pupillometry, and heart rate. We also explored how drivers missing critical events may be a potential new measure. Results varied between focused and mind-wandering mental states and between moving and stopped driving contexts. These observations are an initial step toward understanding mind-wandering across diverse driving scenarios.

Deep neural network (DNN)-based transformer models have demonstrated remarkable performance in natural language processing (NLP) applications. Unfortunately, the unique scaled dot-product attention mechanism and intensive memory access pose a significant challenge during inference on power-constrained edge devices. One emerging solution to this challenge is computing-in-memory (CIM), which uses memory cells for logic computation to reduce data movement and overcome the memory wall. However, existing CIM designs do not support high-precision computations, such as floating-point operations, which are essential for NLP applications. Furthermore, CIM architectures require complex control modules and costly peripheral circuits to harness the full potential of in-memory computation. Hence, this article proposes a scalable RRAM-based in-memory floating-point computation architecture (RIME) that uses single-cycle NOR, NAND, and minority logic to implement in-memory floating-point operations. RIME features efficient parallel and pipeline capabilities with a centralized control module and a simplified peripheral circuit to eliminate data movement during computation. Furthermore, the article proposes pipelined implementations of matrix–matrix multiplication (MatMul) and softmax functions, enabling the construction of a transformer accelerator based on RIME. Extensive experimental results show that compared with GPU-based implementation, the RIME-based transformer accelerator improves timing efficiency by$2.3\times $and energy efficiency by$1.7\times $without compromising inference accuracy.

We address the problem of active mapping with a continually-learned neural scene representation, namely Active Neural Mapping. The key lies in actively finding the target space to be explored with efficient agent movement, thus minimizing the map uncertainty on-the-fly within a previously unseen environment. In this paper, we examine the weight space of the continually-learned neural field, and show empirically that the neural variability, the prediction robustness against random weight perturbation, can be directly utilized to measure the instant uncertainty of the neural map. Together with the continuous geometric information inherited in the neural map, the agent can be guided to find a traversable path to gradually gain knowledge of the environment. We present for the first time an online active mapping system with a coordinate-based implicit neural representation. Experiments in the visually-realistic Gibson and Matterport3D environment demonstrate the efficacy of the proposed method.

Contingency research to find optimal operations and postcontingency recovery plans in distribution networks has gained major attention in recent years. To this end, we consider a multiperiod optimal power flow problem in distribution networks, subject to the N – 1 contingency in which a line or distributed energy resource fails. The contingency can be modeled as a stochastic disruption, an event with random magnitude and timing. Assuming a specific recovery time, we formulate a multistage stochastic convex program and develop a decomposition algorithm based on stochastic dual dynamic programming. Realistic modeling features, such as linearized AC power flow physics, engineering limits, and battery devices with realistic efficiency curves, are incorporated. We present extensive computational tests to show the efficiency of our decomposition algorithm and out-of-samplex performance of our solution compared with its deterministic counterpart. Operational insights on battery utilization, component hardening, and length of recovery phase are obtained by performing analyses from stochastic disruption-aware solutions. Summary of Contribution: Stochastic disruptions are random in time and can significantly alter the operating status of a distribution power network. Most of the previous research focuses on the magnitude aspect with a fixed set of time points in which randomness is observed. Our paper provides a novel multistage stochastic programming model for stochastic disruptions, considering both the uncertainty in timing and magnitude. We propose a computationally efficient cutting-plane method to solve this large-scale model and prove the theoretical convergence of such a decomposition algorithm. We present computational results to substantiate and demonstrate the theoretical convergence and provide operational insights into how making infrastructure investments can hedge against stochastic disruptions via sensitivity analyses.

We consider a robust optimization problem in an electric power system under uncertain demand and availability of renewable energy resources. Solving the deterministic alternating current (AC) optimal power flow (ACOPF) problem has been considered challenging since the 1960s due to its nonconvexity. Linear approximation of the AC power flow system sees pervasive use, but does not guarantee a physically feasible system configuration. In recent years, various convex relaxation schemes for the ACOPF problem have been investigated, and under some assumptions, a physically feasible solution can be recovered. Based on these convex relaxations, we construct a robust convex optimization problem with recourse to solve for optimal controllable injections (fossil fuel, nuclear, etc.) in electric power systems under uncertainty (renewable energy generation, demand fluctuation, etc.). We propose a cutting-plane method to solve this robust optimization problem, and we establish convergence and other desirable properties. Experimental results indicate that our robust convex relaxation of the ACOPF problem can provide a tight lower bound.