Research

Reaction-time variability (RTV) is a Renowned indicator of ADHD, but we still do not fully understand its cognitive roots in adults. We analyzed Continuous Performance Task (CPT-II) data from 99 adults from the HYPERAKTIV dataset (50 ADHD, 49 controls). Analysis combined behavioral analysis, drift diffusion modeling (DDM), and interpretable machine-learning classification in a single framework. Average reaction time did not differ between groups (0.39s vs. 0.39s, p = 0.198), but RTV was significantly higher in the ADHD group (0.130s vs. 0.098s, p = 0.0013, d = 0.62), which indicates attentional instability rather than overall slowing. DDM decomposed this instability into two distinct mechanisms: ADHD participants showed lower drift rates (4.50 vs. 4.78, p = 0.0285, d = −0.447), which reflects less efficient evidence accumulation, and shorter non-decision time (0.146 s vs. 0.171 s, p < 0.001, d = −0.565), suggesting faster response initiation and boundary separation did not differ. Drift rate was correlated with childhood symptom severity on the WURS after FDR correction (r = −0.302, p = 0.016), and non-decision time was the only DDM parameter that independently predicted diagnosis (β = −14.27, p = 0.011). Classification primarily depends on behavioral features (SVM-RBF: ROC-AUC 0.784, 74.2% accuracy), which outperformed DDM-only models (best ) ROC-AUC 0.694, 72.2% accuracy). A Random Forest model combining behavioral and DDM features achieved the highest ROC-AUC overall (0.808), but at lower accuracy (71.1%). A comprehensive search across all 511 possible feature subsets identified a four-feature model (mean reaction time, omission rate, commission rate, and DDM boundary separation) that exceeded the full nine-feature SVM-RBF model on every metric (ROC-AUC 0.802, 79.4% accuracy) while approaching the Random Forest’s top ROC-AUC, including less than half of the predictors. These findings show that adult ADHD variability arises from dissociable evidence-accumulation and response-initiation processes, and that integrating DDM parameters with behavioral data sharpens both mechanistic understanding and diagnostic discrimination.

Current urban drainage systems are under increasing strain due to rapid urbanization, resulting to frequent blockages, urban flooding, and massive public health hazards. To address these issues, we design an event-driven edge intelligence framework that employs spiking neural networks (SNNs) for early detection of blockage-related anomalies in sewer pipelines. Unlike traditional IoT monitoring approaches that rely on centralized processing and continuous data sampling, the proposed system executes local and spike-based inference at distributed sensor nodes. With the help of environmental and physical sensor signals, we collect water-quality field parameters such as conductivity, turbidity, and flow and encode them into spike events by capturing significant deviations from baseline conditions. These events are processed by lightweight SNNs to classify pipeline conditions into four different states: (i) normal; (ii) contaminated; (iii) partial blockage, and (iv) severe blockage. To demonstrate the practicality of our method, we implement a lightweight neuromorphic inference testbed using ESP32 microcontroller with multi-sensor feature fusion approach. Experimental results show that our proposed framework achieves an overall accuracy of 86.60%, with balanced precision (86.40%) and a F1- score (86.43%), demonstrating reliable and efficient multi-class anomaly detection. 

This is the era of data-intensive artificial intelligence in which conventional computing architectures are increasingly constrained by energy consumption and memory bandwidth limitations, which hinder scalability and sustained innovation. Neuromorphic computing, inspired by the remarkable efficiency of biological neural systems, is evolving with the integration of principles from computational neuroscience. This review surveys computational neuroscience inspired neuromorphic computing, covering key methods, biological inspiration, neuromorphic hardware architectures, applications, and emerging technologies developed to address challenges in energy efficiency, scalability, and realtime intelligent processing.