Abstract Ensuring the safety of reinforcement learning (RL) policies in high-stakes environments requires more than formal verification: it needs interpretability and targeted falsification—the deliberate search for counter-examples that expose potential failures before deployment. We present AEGIS-RL (Abstract, Explainable Graphs for Integrated Safety in RL), a hybrid framework that unifies (1) explainable RL, (2) probabilistic model checking, and (3) risk-guided falsification, and augments them with (4) a lightweight runtime safety shield that switches to a fallback policy when estimated risk exceeds a threshold. AEGIS-RL first builds a directed, semantically meaningful graph from offline trajectories that blends local and global explanations to make policy behavior transparent and verifier-friendly. This abstract graph is fed to a probabilistic model checker (e.g., Storm) to verify temporal safety specifications; when violations exist, the checker returns interpretable counterexample traces that pinpoint how the policy fails. When specifications appear satisfied, AEGIS-RL estimates residual risk during checking to steer falsification toward high-risk, under-explored states, broadening coverage beyond the offline data. Across safety-critical benchmarks including two MuJoCo tasks and a medical insulin-dosing scenario; AEGIS-RL uncovers significantly more violations than uncertainty- and fuzzing-based baselines and yields a broader, more novel set of failure trajectories. The resulting explanations and counterexamples provide actionable guidance to understand, debug, and repair unsafe policies while enabling runtime mitigation without retraining.

This paper explores the use of large language models (LLMs) to score and explain short-answer assessments in K-12 science. While existing methods can score more structured math and computer science assessments, they often do not provide explanations for the scores. Our study focuses on employing GPT-4 for automated assessment in middle school Earth Science, combining few-shot and active learning with chain-of-thought reasoning. Using a human-in-the-loop approach, we successfully score and provide meaningful explanations for formative assessment responses. A systematic analysis of our method's pros and cons sheds light on the potential for human-in-the-loop techniques to enhance automated grading for open-ended science assessments.

Given a corpus of documents, focused analysis aims to find topics relevant to aspects that a user is interested in. The aspects are often expressed by a set of keywords provided by the user. Short texts such as microblogs and tweets pose several challenges to this task because the sparsity of word co-occurrences may hinder the extraction of meaningful and relevant topics. Moreover, most of the existing topic models perform a full corpus analysis that treats all topics equally, which may make the learned topics not be on target. In this paper, we propose a novel targeted topic model for semantic short-text embedding which aims to learn all topics and low-dimensional visual representations of documents, while preserving relevant topics for focused analysis of short texts. To preserve the relevant topics in the visualization space, we propose jointly modeling topics and the pairwise document ranking based on document-keyword distances in the visualization space. The extensive experiments on several real-world datasets demonstrate the effectiveness of our proposed model in terms of targeted topic modeling and visualization.

Deep generative diffusion models are a promising avenue for 3D de novo molecular design in materials science and drug discovery. However, their utility is still limited by suboptimal performance on large molecular structures and limited training data. To address this gap, we explore the design space of E(3)-equivariant diffusion models, focusing on previously unexplored areas. Our extensive comparative analysis evaluates the interplay between continuous and discrete state spaces. From this investigation, we present the EQGAT-diff model, which consistently outperforms established models for the QM9 and GEOM-Drugs datasets. Significantly, EQGAT-diff takes continuous atom positions, while chemical elements and bond types are categorical and uses time-dependent loss weighting, substantially increasing training convergence, the quality of generated samples, and inference time. We also showcase that including chemically motivated additional features like hybridization states in the diffusion process enhances the validity of generated molecules. To further strengthen the applicability of diffusion models to limited training data, we investigate the transferability of EQGAT-diff trained on the large PubChem3D dataset with implicit hydrogen atoms to target different data distributions. Fine-tuning EQGAT-diff for just a few iterations shows an efficient distribution shift, further improving performance throughout data sets. Finally, we test our model on the Crossdocked data set for structure-based de novo ligand generation, underlining the importance of our findings showing state-of-the-art performance on Vina docking scores.

It is common to listen to songs that match one's mood. Thus, an AI music recommendation system that is aware of the user's emotions is likely to provide a superior user experience to one that is unaware. In this paper, we present an emotion-aware music recommendation system. Multiple models are discussed and evaluated for affect identification from a live image of the user. We propose two models: DRViT, which applies dynamic routing to vision transformers, and InvNet50, which uses involution. All considered models are trained and evaluated on the AffectNet dataset. Each model outputs the user's estimated valence and arousal under the circumplex model of affect. These values are compared to the valence and arousal values for songs in a Spotify dataset, and the top-five closest-matching songs are presented to the user. Experimental results of the models and user testing are presented.

Equivariant neural networks, whose hidden features transform according to representations of a group $G$ acting on the data, exhibit training efficiency and an improved generalisation performance. In this work, we extend group invariant and equivariant representation learning to the field of unsupervised deep learning. We propose a general learning strategy based on an encoder-decoder framework in which the latent representation is separated in an invariant term and an equivariant group action component. The key idea is that the network learns to encode and decode data to and from a group-invariant representation by additionally learning to predict the appropriate group action to align input and output pose to solve the reconstruction task. We derive the necessary conditions on the equivariant encoder, and we present a construction valid for any $G$, both discrete and continuous. We describe explicitly our construction for rotations, translations and permutations. We test the validity and the robustness of our approach in a variety of experiments with diverse data types employing different network architectures.

Rapid prototyping makes additive manufacturing (or 3D printing) useful in critical application domains such as aerospace, automotive, and medical. The rapid expansion of these applications should prompt the examination of the underlying security of 3D printed objects. In this paper, we present Mystique, a novel class of stealthy attacks on printed objects that leverage the fourth dimension of emerging 4D printing technology to introduce embedded logic bombs through manufacturing process manipulation. Mystique enables visually benign objects to behave maliciously upon the activation of the logic bomb during operation. It leverages the manufacturing process to embed a physical logic bomb that can be triggered with specific stimuli to change the physical and mechanical properties of the printed objects. These changes in properties can potentially cause catastrophic operational failures when the objects are used in critical applications such as drones, prosthesis, or medical applications.

Bridge condition assessment is important to maintain the quality of highway roads for public transport. Bridge deterioration with time is inevitable due to aging material, environmental wear and in some cases, inadequate maintenance. Non-destructive evaluation (NDE) methods are preferred for condition assessment for bridges, concrete buildings, and other civil structures. Some examples of NDE methods are ground penetrating radar (GPR), acoustic emission, and electrical resistivity (ER). NDE methods provide the ability to inspect a structure without causing any damage to the structure in the process. In addition, NDE methods typically cost less than other methods, since they do not require inspection sites to be evacuated prior to inspection, which greatly reduces the cost of safety related issues during the inspection process. In this paper, an autonomous robotic system equipped with three different NDE sensors is presented. The system employs GPR, ER, and a camera for data collection. The system is capable of performing real-time, cost-effective bridge deck inspection, and is comprised of a mechanical robot design and machine learning and pattern recognition methods for automated steel rebar picking to provide realtime condition maps of the corrosive deck environments.

Satisfactory operation of civil infrastructure is of critical importance to an economy. In order to maintain performance, infrastructure needs to be properly maintained. Inspecting infrastructure is inherently labor-intensive work and costly. In this paper, we propose a solution to cost-effective infrastructure inspection by developing a multi-functional inspection robot. The robot is equipped with several state-of-the-art non-destructive evaluation (NDE) sensors to perform inspection. The robot is able to perform selected inspection methods in certain areas based on multiple sensor data fusion. With this, the overall inspection time is reduced, which in turn reduces maintenance cost. An inspection framework based on multiple NDE data sensor fusion is proposed. Detailed discussions include robot design, robot navigation and sensor data fusion.

A low-power wearable wireless sensor measuring both molecular biomarkers and physiological signals is proposed, where the former are measured by a microfluidic biosensing system while the latter are measured electrically. The low-power consumption of the sensor is achieved by an all-analog circuit implementing Analog Joint Source-Channel Coding (AJSCC) compression. The sensor is applicable to a wide range of biomedical applications that require real-time concurrent molecular biomarker and physiological signal monitoring.

Delay Tolerant Networks (DTNs) are sparse mobile ad-hoc networks in which there is typically no complete path between the source and destination. Although many routing algorithms for DTNs have been proposed, prior works generally focus on utilizing the delivery probability of network nodes and the social network structure for data forwarding. In this work, we investigate the use of the inter-contact time (ICT) distribution to derive a new metric that selects the next relay node with the least expected minimum delay (EMD) among all possible routes to the destination. We address the case of exponential and power-law ICTs, which are the most popular assumptions for ICTs that have emerged in recent literature. Extensive simulation results based on the Cabspotting and Cambridge Haggle traces show that our proposed metric can achieve up to 21% higher delivery rate and 23% lower delay than existing schemes.

Delay Tolerant Networks (DTNs) are sparse mobile ad-hoc networks in which there is typically no complete path between the source and destination. While much work has been done in the design of forwarding algorithms, little work has focused on studying forwarding under the presence of short contact durations. In this paper, we study a single- copy contact duration-aware (CDA) routing strategy. We address two key issues: (1) to which next hop relay node should messages be forwarded and (2) in which order should messages be forwarded. To reduce the transmission cost, we select relay nodes from both current and past contacts based on the one-hop and two-hop delivery probability, respectively. We derive the delivery probability from the distribution of contact duration time and inter- contact time. For the message scheduling, messages with the highest delivery probability are prioritized to be transmitted first. Extensive simulation results based on the Cabspotting trace show that our scheme can achieve up to 13% higher delivery rate, 12% lower delay, and 23% lower transmission cost compared to other routing strategies.

Among the major challenges in the realization of practical health monitoring systems is the identification of short-duration events from larger signals. Time-series segmentation refers to the challenge of subdividing a continuous stream of data into discrete windows, which are individually processed using statistical classifiers to recognize various activities or events. In this paper, we propose a probabilistic algorithm for segmenting time-series signals, in which window boundaries are dynamically adjusted when the probability of correct classification is low. Our proposed scheme is benchmarked using an audio-based nutrition-monitoring case-study. Our evaluation shows that algorithm improves the number of correctly classified instances from a baseline of 75% to 94% using the RandomForest classifier.

Trustworthy and usable healthcare requires not only effective disease diagnostic procedures to ensure delivery of rapid and accurate outcomes, but also lightweight user privacy-preserving capabilities for resource-limited medical sensing devices. In this paper, we present MedSen, a portable, inexpensive and secure smartphone-based biomarker1 detection sensor to provide users with easy-to-use real-time disease diagnostic capabilities without the need for in-person clinical visits. To minimize the deployment cost and size without sacrificing the diagnostic accuracy, security and time requirement, MedSen operates as a dongle to the user's smartphone and leverages the smartphone's computational capabilities for its real-time data processing. From the security viewpoint, MedSen introduces a new hardware-level trusted sensing framework, built in the sensor, to encrypt measured analog signals related to cell counting in the patient's blood sample, at the data acquisition point. To protect the user privacy, MedSen's in-sensor encryption scheme conceals the user's private information before sending them out for cloud-based medical diagnostics analysis. The analysis outcomes are sent back to Med-Sen for decryption and user notifications. Additionally, MedSen introduces cyto-coded passwords to authenticate the user to the cloud server without the need for explicit screen password entry. Each user's password constitutes a predetermined number of synthetic beads with different dielectric characteristics. MedSen mixes the password beads with the user's blood before submitting the data for diagnostics analysis. The cloud server authenticates the user based on the statistics and characteristics of the beads with the blood sample, and links the user's identity to the encrypted analysis outcomes. We have implemented a real-world working prototype of MedSen through bio-sensor fabrication and smartphone app (Android) implementations. Our results show that MedSen can reliably classify different users based on their cyto-coded passwords with high accuracy. MedSen's built-in analog signal encryption guarantees the user's privacy by considering the smartphone and cloud server possibly untrusted (curious but honest). MedSen's end-to-end time requirement for disease diagnostics is approximately 0.2 seconds on average.

In Delay Tolerant Networks (DTNs) with resource constraints such as short contact durations and small buffers, message scheduling and drop prioritization is a critical issue as it affects the routing performance. Current solutions mainly focus on devising buffer management strategies by assuming that the contact rates between mobile nodes are exponentially distributed. While this assumption is suitable for vehicular mobility scenarios such as taxicabs in a city, it is often invalid for mobility traces that feature human-assisted devices. Recent studies have shown that human mobility traces follow a truncated power-law distribution. In this paper, we propose a new buffer management strategy based on power-law distributed contacts. The main objective is to minimize the average message delivery delay in DTN networks with resource constraints and heterogeneous node mobility. We focus on two key issues: (1) in which order should messages be replicated when contact duration and forwarding bandwidth are limited, and (2) which messages should be dropped first when the buffer is full. We develop a utility function using global network information to compute per-packet average delay utility. Messages are then scheduled and dropped according to their utility values. Extensive simulation results based on real-life human mobility traces show that our proposed scheme can deliver messages in up to 27% less time than existing schemes, while still achieving a high delivery ratio.

In this paper, we address several security issues in our previously proposed content retrieval scheme for Disruption Tolerant Networks (DTNs). The content retrieval is built upon the social-tie relationships among DTN nodes for routing and content lookup service placement. Malicious nodes can launch attacks by advertising falsified social-tie information to attract and drop packets intended for other nodes, or simply disrupt and destroy the query and delivery paths. Furthermore, selfish nodes, while not seeking to attack, are unwilling to forward packets of others. Both malicious and selfish behaviors contribute to the deterioration of the content retrieval performance. To address the problem, we propose to secure both social-tie records and content delivery records during a contact between two nodes. The unforgeable social-tie records prevent malicious nodes from falsifying the social-tie information. The delivery records from which the packet forwarding ratio of a node is computed, help detect selfish behavior. Lastly, we propose a blacklist distribution method that allows nodes to filter out misbehaving nodes from their social contact graph, effectively preventing network traffic from flowing to misbehaving nodes. Extensive real-trace-driven simulation results show that our scheme can detect misbehaving nodes and mitigate their effects efficiently, thus improving the content retrieval performance.

Due to the unstable network topology of Delay Tolerant Networks (DTNs), multi-copy routing is often used to increase the reliability of message delivery. However, this routing approach suffers from high buffer and bandwidth overhead. While much work has been done in the design of forwarding algorithms, little work has focused on studying forwarding under the presence of resource constraints such as short contact durations and small buffers. In this paper, we investigate a multi-copy routing strategy and a buffer management policy that maximize the delivery rate in DTNs. We consider a realistic DTN environment with resource constraints, heterogeneous node mobility, and varied message sizes. There are three key issues in DTN routing: (1) to which next hop relay node should messages be replicated, (2) in which order should messages be replicated, and (3) which messages should be dropped first when the buffer is full. We propose to forward a message to a neighboring node that has both a stronger social tie with the destination and a smaller or similar queue length. This aims to reduce traffic at highly connected network nodes, avoiding frequent message drops which compromise the delivery ratio. For the second and third issue, we develop a utility function using global network information to compute per-packet delivery rate utility. Messages are then scheduled and dropped according to their utility values. Extensive simulation results based on the real-world San Francisco cab trace show that our proposed scheme can achieve a delivery rate of up to 22% higher than existing schemes, while still maintaining a comparable average delivery delay. Furthermore, our scheme distributes the network loads more evenly, with the top 10% of network nodes handling only 24% of the forwardings.

Abstract Delay‐tolerant networks are sparse mobile ad hoc networks in which there is typically no complete path between the source and destination. Although many routing schemes for delay‐tolerant networks have been proposed, they do not address fairness issues in terms of the workload/traffic handled at each node and the share of throughput among different destination nodes. In this paper, we propose a socially aware routing strategy that optimizes both fairness and throughput. A relay node is selected based on the multi‐hop delivery probability and its queue length. The effect of queue length control is to divert traffic away from highly connected nodes and allows nodes to explore less‐congested paths to the destination. This helps balance the network loads, thus achieving workload fairness. Furthermore, to achieve throughput fairness, we sort arriving messages into different destination‐based queues. Messages are then scheduled following a two‐level forwarding strategy that optimizes throughput fairness using round‐robin and delivery ratio using priority scheduling. Extensive real‐trace‐driven simulation results show that our scheme outperforms existing algorithms in terms of the delivery ratio. Furthermore, our scheme achieves a high throughput fairness, while distributing the network loads more evenly, with the top 10% of network nodes handling only 22% of the forwardings. Copyright © 2016 John Wiley & Sons, Ltd.

Poor adherence to prescription medication can compromise treatment effectiveness and cost the billions of dollars in unnecessary health care expenses. Though various interventions have been proposed for estimating adherence rates, few have been shown to be effective. Digital systems are capable of estimating adherence without extensive user involvement and can potentially provide higher accuracy with lower user burden than manual methods. In this paper, we propose a smartwatch-based system for detecting adherence to prescription medication based the identification of several motions using the built-in tri-axial accelerometers and gyroscopes. The efficacy of the proposed technique is confirmed through a survey of medication ingestion habits and experimental results on movement classification.

Delay Tolerant Networks (DTNs) are sparse mobile ad-hoc networks in which there is typically no complete path between the source and the destination. Data routing in DTNs is challenging, and has attracted much attention from the research community. Although many routing schemes have been proposed, they do not address the fairness issue in data delivery to different destination nodes. In this paper, we propose a novel socially-aware routing strategy that optimizes both fairness and throughput. We combine controlled data “spraying” at the source node and single-copy routing at the intermediate nodes. A replication decision is made based on the delivery probability computed over the most probable path in the social contact graph. Furthermore, at intermediate nodes, we sort arriving data into different queues. We then propose a two-level data forwarding strategy that optimizes fairness using round-robin at the first level and throughput using priority scheduling at the second level. Through extensive simulation studies using a real-world mobility trace, we show that our scheme achieves a high delivery ratio, low delay, and low replication overhead.

In this paper, we propose a distributed content retrieval scheme for Disruption Tolerant Networks (DTNs). Our scheme consists of two key components: a content discovery (lookup) service and a routing protocol for message delivery. Both components rely on three key social metrics: centrality, social level, and social tie. Centrality guides the placement of the content lookup service. Social level guides the forwarding of content requests to a content lookup service node. Social tie is exploited to deliver content requests to the content provider, and content data to the requester node. We leverage bounded random walks to estimate a node's centrality. The X-means clustering algorithm is used to compute a node's social level. Lastly, a node's social tie is computed based on the frequency and recency of node contacts. Extensive real-trace-driven simulation results show that our scheme requires less control overhead while maintaining comparable performance for content retrieval applications.