2025-12-10 –, Deborah Sampson
Machine learning often assumes clean, high-quality data. Yet the real world is noisy, incomplete, and messy, and models trained only on sanitized datasets become brittle. This talk explores the counterintuitive idea that deliberately corrupting data during training can make models more robust. By adding structured noise, masking inputs, or flipping labels, we can prevent overfitting, improve generalization, and build systems that survive real world conditions. Attendees will leave with a clear understanding of why “bad data” can sometimes lead to better models.
This talk examines the role of data corruption as regularization in machine learning pipelines. The objective is to show that structured noise, far from being a nuisance, can strengthen models against distribution shifts, adversarial examples, and missing data. The session begins with motivation: real-world signals are never clean, and models trained only on perfect datasets fail in practice. We then introduce different types of corruption: additive noise such as Gaussian or salt and pepper, occlusion and masking for images and text, and controlled label flipping for stress testing. Each will be demonstrated with examples in Python, showing how corruption acts as a training signal that forces models to learn deeper patterns. Case studies from computer vision and NLP highlight how corrupted inputs improve resilience to noisy or adversarial environments.
The audience is intermediate level data scientists and engineers familiar with supervised learning and Python ML libraries. No advanced math is required. The talk will be informative and applied, balancing intuition, visual examples, and code demonstrations. By the end, attendees will understand how to incorporate data corruption deliberately into training pipelines, when it helps, when it hurts, and why noise can be a hidden ally in building robust machine learning systems.
Aayush Gauba is a researcher and developer working at the intersection of machine learning, quantum-inspired models, and AI security. He has created open-source projects such as AIWAF, an adaptive web application firewall, and has published research on quantum-inspired neural architectures and robust learning methods. His work focuses on building practical tools that are both scientifically innovative and accessible to the wider Python community. Outside of research, Aayush is passionate about sharing knowledge through talks, tutorials, and collaborations that bridge theory with real world application.