Security of Neural Networks
Defending deep models against adversarial examples, backdoor and data-poisoning attacks — and understanding when and why they fail.
Research studies, software and prototypes from across the NPerseus team.
Research studies, prototypes and software described in public profiles. Status, roles and funding follow each source record; a profile entry does not by itself establish a released product.
Applications, tools, plugins and prototypes members have built. Engineering evidence, kept apart from the research record.
42 matching software project records · Page 3 of 4
public TensorFlow/Keras notebook implementing ResUNet segmentation with focal Tversky loss.
explored binary translation and compiler tooling using Rust and LLVM.
built and maintains a full-stack research-group platform for member profiles, publications, projects and collaboration workflows.
trained and evaluated a transfer-learning CNN for chest X-ray classification.
experimental PlayStation 2 binary translator built with Rust and LLVM tooling.
built a MERN application that uses CNN, MobileNetV2 and ResNet models to classify sketches from the Google QuickDraw dataset.
Java Mindustry plugin for automatic item and liquid replenishment using cached building requirements; developed from an earlier JavaScript replenishment mod, with RapidNew and Radpid-MindustryPlugin counted as one project family.
built a web application for discovering and managing recipes.
Java/Spring Boot prototype for food-product APIs, QR-code generation and basic payment-transaction records, using configurable H2/MySQL storage.
a Python/Ollama prototype that checks syntax, generates code corrections and retries following execution errors.
a policy-document question-answering prototype combining risk detection, adaptive retrieval, answer evaluation, evidence checks and retrieval repair.
React and Material UI interview-management prototype with interview creation, expert listings, search and a skill-discussion interface.
Defending deep models against adversarial examples, backdoor and data-poisoning attacks — and understanding when and why they fail.
Reversible data hiding in encrypted images, image and text steganography, and robust watermarking for ownership and integrity.
Detecting manipulated and GAN-generated media — deepfakes, splicing and median-filtering traces.
Federated and privacy-preserving learning, explainable and trustworthy models, and methods for limiting the disclosure of training data.