A deployed agent rarely breaks all at once. A model update, a changed tool, or a shift in what users ask for can raise its violation rate a little at a time, and nobody notices until the damage is done. We ask how quickly different change detectors notice such drift when each is set to the same false-alarm rate. We compare a CUSUM, a likelihood-ratio CUSUM, an EWMA chart, a rolling-mean threshold, a sliding-window test and an ADWIN-style adaptive window on streams of per-task violation indicators, with every detector calibrated to one false alarm per 1,000 tasks and checked on fresh data. For a step increase from 5% to 10%, the CUSUM family, the EWMA chart and the adaptive window all needed about 69 to 75 tasks (median), the rolling mean 79, and the sliding-window test 120. On bursty drift the gap widens to 4.8 times. For a slow ramp, the classical first-order delay formula was wrong by up to 9 times. We derive a better one, treating the CUSUM as a reflected random walk with slowly changing drift, and it predicts the simulated median delay within 8% over a hundredfold range of slopes. Estimating the baseline from 500 tasks raised false alarms by up to 1.39 times for detectors that use it, and benign seasonality of 100% amplitude raised them about threefold. All detectors ran in under 2 µs per observation. All results are simulation-based.

Keywords : Drift Detection, CUSUM, EWMA, ADWIN, Agent Monitoring, False-Alarm Calibration.

Authors : Appiah Prince , Vijayalakshmi Kakulapati

Title : Noticing the Slide : Calibrated Drift Detection for Deployed Agents

Volume/Issue : 2026;3(4 (October-December))

Page No : 13 - 22

Optical vision is a crucial part of self-driving cars. Vehicle, street, building, pedestrian, and road sign detection that is accurate could help self-driving cars drive as safely as human beings. Object detection, on the other hand, has been a tough task for years due to the effects of lighting, topology, and occlusion on photographs of objects in the real world. Many Convolutional Neural Networks (CNN) based algorithms have improved the detection performance in larger samples in recent years. These approaches have sluggish identification speed, on the other hand, limit their use in real-time scenarios. The Resnet50, revolutionary network architecture with exceptional efficiency acted as the backbone network and the addition of CAR (Concatenation and residual) architecture in the YOLO model is proposed in this research. This architecture uses a topology similar to that of a residual network, but with a few more layers and a few skip connections between layers for which feature reuse is not necessary. This enhanced network is used to detect real-time objects in the You Look Only Once version 3 (YOLO V3) detector. COCO, ImageNet, and PASCAL VOC databases as well as custom datasets are used to train this network for Indian Vehicles. The proposed network outperforms state-of-the-art CNNs, according to the results of the experiments.

Keywords : Deep Learning, CNN , Resnet50, Self-Driving Vehicles, Skip Connections, YOLOv3.

Authors : Venkata Krishna Reddy Atchi

Title : An Improved YoLov3 Framework For Real-Time Object Detection In Autonomous Vehicles

Volume/Issue : 2026;3(4 (October-December))

Page No : 1 - 12