Inter-satellite link (ISL) spoofing in Low Earth Orbit (LEO) constellations poses a growing yet underexplored security challenge. This letter presents EMMA, a lightweight Multi-task Learning (MTL)-based framework for ISL authentication and spoof detection using simple physical behavior-based features. It also presents the first benchmark simulation dataset on ISL security. EMMA achieves detection rates of 94.75% and 94.46% on registered and spoofing satellite samples, respectively, at low inference latency ( 10(-4) seconds) and low memory overhead (3.5 KB). We also compare EMMA with existing works to demonstrate its comparative effectiveness.
Dyslexia screening remains a challenging clinical problem due to its multimodal cognitive manifestations, reliance on specialist assessment, and limited scalability in conventional healthcare settings. While the Intelligent Internet of Medical Things (IIoMT) enables data-driven dyslexic pattern assessment through distributed sensing, existing solutions often lack adaptive intelligence, contextual continuity, and integrated security. This article presents a secure multi-agentic hybrid IIoMT framework for multimodal dyslexia screening, combining agent-to-agent (A2A) intelligence, Model Context Protocol (MCP)-based orchestration, and longitudinal cognitive reasoning. The proposed framework integrates four complementary screening tasks—face drawing, clock drawing, handwriting analysis, and reading assessment with gaze tracking—captured via heterogeneous IIoMT endpoints. Specialized AI agents analyze each modality and collaborate through A2A coordination to enable multimodal feature fusion and unified dyslexia risk assessment. The use of persistent contextual intelligence is a key innovation, enabling the retention of clinically relevant interaction context across sessions for personalized and progressive screening. Security and privacy are embedded as system properties, addressing data protection, multi-agentic AI model trustworthiness, and secure, ethical, and authenticated inter-agent collaboration. Experimental evaluation on dyslexia screening demonstrates achieving 94.5% accuracy in face drawing analysis, 82.7% in clock drawing, 96.7% in handwriting assessment, and 83.3% accuracy in reading fluency detection (n = 120). The proposed framework advances dyslexia screening toward scalable, intelligent, and longitudinal digital assessment.
Automated short-answer grading research has historically focused on exams consisting solely of questions pertaining to a single topic. Automatic grading of exams containing questions about more than one topic remains less explored. In this work, a Graph-Retrieval Automated Scoring Pipeline (GRASP) is introduced for grading label-free multi-topic science exams. Label-free exams are short-answer exams in which a student's responses to several distinct topics are merged into a single paragraph, with no markup labels or segmentation indicating which span answers which question. Reference answers for each question are encoded into a FAISS vector index via Sentence-BERT, and a semantic similarity graph is constructed over this set of reference answers. At grading time, sentence count heuristics, with a large language model used to resolve ambiguous cases, are first applied to predict how many distinct topics were answered in the student essay. This process is performed without training data or domain-specific example essays. Candidate reference nodes, each storing one (question, reference answer, concatenation of both) from the reference index, are then retrieved through cosine similarity based Retrieval-Augmented Generation (RAG) and Graph Retrieval-Augmented Generation (GRAG). GRAG operates by taking the top cosine matches as seed nodes and then performing a graph traversal over strong edges to find additional reference nodes that may have been missed by RAG. The Hungarian algorithm is then used to optimally assign one reference node per question segment such that no reference is duplicated. Each segment is then graded against its assigned reference independently using GPT-4.1-mini. This experiment is performed to show the effect of retrieval quality on grading accuracy and the benefit of graph-augmented retrieval versus strict cosine similarity methods at various levels of essay complexity.
We address a practical failure mode in cameraLiDAR fusion for urban driving: extremely small image disparities that, when converted directly to metric depth, yield unstable estimates and derail the front end before LiDAR cues can anchor the pose. We propose a minimal, intrinsics-aware conditioning step prior to fusion that rescales disparity using the true focal length and baseline and enforces a small minimumdisparity floor to bound depth, complemented by conservative dynamic-object masking and a LiDAR-anchored back end. On representative sequences, we verify that the data operate in a “small-disparity” regime and show that conditioning restores usable depth maps and allows the fusion pipeline to proceed. We further report preliminary trajectory metrics on KITTI Odometry Sequence 00 using standard absolute and relative pose errors: a LiDAR ICP baseline (filtered point-to-plane) achieves APE RMSE $28.14 \mathbf{m}$ and RPE ($\Delta=10$) translation/rotation RMSE $17.37 \mathbf{m} / 23.58^{\circ}$. A disparity-conditioned point-selection variant yields comparable $\operatorname{RPE}(\Delta=10)\left(17.04 \mathbf{~ m} / 23.42^{\circ}\right)$ but higher APE RMSE ($46.67 \mathbf{m}$), suggesting that disparity-based gating alone can reduce long-range geometric constraints needed for global consistency. These results quantify end-to-end operability and highlight where conditioning must be integrated as an explicit residual/prior (rather than selection-only) to improve trajectory accuracy.
Accurate prediction of waves is crucial for the provision of marine-related activities, such as harbor operations, naval navigation, and numerous coastal and offshore ventures. Dependable forecasts of the wave state are essential to ensure safe operations, streamline maritime logistical procedures, and minimize dangers related to a challenging-to-predict sea state. In this project, four models are implemented: Decision Forest Regression (DFR), Neural Network Regression (NNR), Boosted Decision Trees Regression (BDTR), and Linear Regression (LR). The primary objectives of the project are to (i) implement four different prediction models independently on an ocean wave prediction dataset, and (ii) determine the best prediction model among the four in predicting ocean waves. Six evaluation metrics are used to measure the models’ performance, and the results indicate that BDTR performs better across all evaluation metrics, achieving the highest coefficient of determination (COD) and accuracy values, as well as the lowest error measurements. NNR was the worst, with the lowest COD and the highest errors. DFR and LR models were provided with moderate accuracy.
