Novel Stealth Loss

Novel stealth loss functions are being developed to improve the performance and robustness of various machine learning models, particularly in adversarial settings and continual learning. Research focuses on designing losses that enhance model transferability across different datasets and tasks, while mitigating issues like catastrophic forgetting and maintaining model plasticity. These advancements are significant because they address limitations in existing models, leading to more robust and adaptable AI systems with applications ranging from object detection and autonomous driving to online learning and system identification.

Papers

July 7, 2024

Privacy of the last iterate in cyclically-sampled DP-SGD on nonconvex composite losses
Weiwei Kong, Mónica Ribero
Differential Privacy Privacy Policy Private Stochastic General Nonconvex Differential Privacy SGD Novel Stealth Loss

March 31, 2024

Addressing Loss of Plasticity and Catastrophic Forgetting in Continual Learning
Mohamed Elsayed, A. Rupam Mahmood
Continual LEArning Representation Learning Catastrophic Forgetting Continual Learning Plasticity Rule Novel Stealth Loss

August 14, 2023

ACTIVE: Towards Highly Transferable 3D Physical Camouflage for Universal and Robust Vehicle Evasion
Naufal Suryanto, Yongsu Kim, Harashta Tatimma Larasati, Hyoeun Kang, Thi-Thu-Huong Le, Yoonyoung Hong, Hunmin Yang, Se-Yoon Oh, Howon Kim
Generating User Activity Universal Image Prey Camouflage Adversarial CAmouflage Camouflage Generation Novel Stealth Loss Camouflage Attack

April 6, 2023

Adaptive Decision-Making with Constraints and Dependent Losses: Performance Guarantees and Applications to Online and Nonlinear Identification
Michael Muehlebach
Regret Bound Participation Constraint Expert Model Performance Guarantee Nonlinear System Identification Sub Optimal Adaptive Decision Making Performance Objective Novel Stealth Loss

January 31, 2023

Online Learning in Dynamically Changing Environments
Changlong Wu, Ananth Grama, Wojciech Szpankowski
Online Learning Environment Feature Regret Minimization Regret Analysis Convex Loss Sub Linear Regret Novel Stealth Loss

October 28, 2022

Evaluating the Impact of Loss Function Variation in Deep Learning for Classification
Simon Dräger, Jannik Dunkelau
Deep Learning Deep Neural Network Classification Code Loss Function Global Impact Variational Loss Novel Stealth Loss

Novel Stealth Loss

Papers

Privacy of the last iterate in cyclically-sampled DP-SGD on nonconvex composite losses

Addressing Loss of Plasticity and Catastrophic Forgetting in Continual Learning

ACTIVE: Towards Highly Transferable 3D Physical Camouflage for Universal and Robust Vehicle Evasion

Adaptive Decision-Making with Constraints and Dependent Losses: Performance Guarantees and Applications to Online and Nonlinear Identification

Online Learning in Dynamically Changing Environments

Evaluating the Impact of Loss Function Variation in Deep Learning for Classification