Machine Learning
Face Recognition Evaluation and Detector Robustness
Ensembled face-recognition models across 15.7M IJB-C pairs and reduced detector failures by 46% in a separate robustness evaluation.
20 Projects
Newest FirstMachine Learning
Ensembled face-recognition models across 15.7M IJB-C pairs and reduced detector failures by 46% in a separate robustness evaluation.
Machine Learning
Combined I3D features, Transformer cross-attention, and pose-derived supervision to reduce relative L2 error by 6.46% on FineDiving, with frame-aligned diagnostic evidence.
SoftwareMachine Learning
Processed 20 million MQTT records through a distributed cloud pipeline; a random forest led the six-class model comparison at 78.52% test accuracy.
Software
Cached shortest paths and stabilized live boundaries in a Java segmentation tool, reducing completion time by 5.2–11.2× on two reported tasks.
Machine LearningEmbedded
Heart-rate-aware adaptive inference reached 91.5% accuracy at 1.33 ms on Raspberry Pi 4, with no misses in the 1,000-sample deadline evaluation.
IEEE RTCSA 2025 · Second AuthorRobotics & ControlMachine Learning
Connected language instructions to ROS actions and execution feedback, demonstrating multi-step control on a physical JAKA Zu5 arm.
Robotics & ControlMachine Learning
Built PyTorch agents for swing-up and stabilization, achieving 90% success in the reported noisy single-pendulum simulations.
Machine Learning
A compact CNN reached 99.53% accuracy on a course fundus-image split, with approximately 2.2× faster training than fine-tuned ResNet18.
EmbeddedSoftware
Integrated motion sensing, a matrix display, a servo, and Flask controls into a Raspberry Pi device that responds to taps and tilt.
Robotics & Control
Implemented Nav2 planning and control plugins, connecting autonomous navigation to visual target approach and reachability-aware grasping.
Machine Learning
Implemented clustering and dimensionality reduction from scratch, tracing how initialization, soft assignments, and representation change experimental results.
Machine Learning
Built NumPy classifiers with manual backpropagation, using controlled synthetic experiments to connect architecture choices with decision boundaries and F1.
Robotics & ControlEmbedded
Integrated vision, calibration, and feedback control into a laser-tracking prototype with sub-centimeter error in the reported tests. Guangdong Second Prize.
Robotics & ControlEmbedded
Linked wireless state exchange with onboard flight control to demonstrate coordinated motion between two quadcopters. Third Prize.
Embedded
Inferred mass from a motorized lever balance using filtering and friction compensation, with reported error within ±1% over 75–750 g.
Robotics & ControlEmbedded
Integrated room-number recognition with an autonomous delivery vehicle; expanded training data raised reported recognition accuracy from roughly 80% to over 95%.
Machine Learning
Combined time-series forecasting, neural prediction, and clustering to explain Wordle participation, attempt distributions, and difficulty.
Embedded
Recovered a human-motion Doppler trace from 20 field-recording segments using cellular signals, without a dedicated radar transmitter.
Embedded
Integrated an STM32 smartwatch with FreeRTOS, LVGL, and an embedded Python runtime, supporting script installation over USB.
Embedded
Built a tone-vocoder pipeline and interactive MATLAB console to compare how frequency bands and envelope cues shape reconstructed speech.
Try a different term or browse the full collection.