Benign Input

"Benign input" research explores how seemingly harmless inputs interact with and reveal properties of machine learning models, encompassing both their vulnerabilities and capabilities. Current research focuses on understanding model behavior under benign inputs, including analyzing overfitting phenomena in various architectures like transformers and CNNs, evaluating circuit accuracy, and developing robust fingerprinting techniques. This work is crucial for improving model safety and reliability in applications ranging from medical image analysis to content moderation, ultimately contributing to more trustworthy and secure AI systems.

Papers

September 26, 2024

Benign or Not-Benign Overfitting in Token Selection of Attention Mechanism
Keitaro Sakamoto, Issei Sato
Neural Network Attention Mechanism Benign Overfitting Attention Based Architecture Benign Input Token Selection

July 27, 2024

AResNet-ViT: A Hybrid CNN-Transformer Network for Benign and Malignant Breast Nodule Classification in Ultrasound Images
Xin Zhao, Qianqian Zhu, Jialing Wu
Ultrasound Image Multi Head Self Attention Global Feature Cancer Classification Breast Ultrasound Hybrid CNN Transformer Breast Cancer Classification Benign Input Global Image Feature

July 21, 2024

Adversarial Circuit Evaluation
Niels uit de Bos, Adrià Garriga-Alonso
Adversarial Text Large Scale Circuit Adversarial Evaluation Benign Input

June 28, 2024

Automatic register identification for the open web using multilingual deep learning
Erik Henriksson, Amanda Myntti, Saara Hellström, Anni Eskelinen, Selcen Erten-Johansson, Veronika Laippala
Deep Learning Model Multilingual Model Computational Linguistics Multilingual Corpus Automatic Identification Benign Input

May 24, 2024

ART: Automatic Red-teaming for Text-to-Image Models to Protect Benign Users
Guanlin Li, Kangjie Chen, Shudong Zhang, Jie Zhang, Tianwei Zhang
Adversarial Attack Generative Model Text to Image Model Art Specific Information Red Teaming Pre Trained Generative Model Benign Input

April 22, 2024

Exploring Kinetic Curves Features for the Classification of Benign and Malignant Breast Lesions in DCE-MRI
Zixian Li, Yuming Zhong, Yi Wang
Classification Code Breast Cancer Benign Input Dce Mri Breast Lesion

October 26, 2023

X-SNS: Cross-Lingual Transfer Prediction through Sub-Network Similarity
Taejun Yun, Jinhyeon Kim, Deokyeong Kang, Seong Hoon Lim, Jihoon Kim, Taeuk Kim
Cross Lingual Transfer Multilingual Language Model Model Adaptation Benign Input

May 24, 2023

Malicious or Benign? Towards Effective Content Moderation for Children's Videos
Syed Hammad Ahmed, Muhammad Junaid Khan, H. M. Umer Qaisar, Gita Sukthankar
Gameplay Video Nine Year Old Child Content Moderation Malicious Content Benign Input Malicious Data

March 25, 2023

No more Reviewer #2: Subverting Automatic Paper-Reviewer Assignment using Adversarial Learning
Thorsten Eisenhofer, Erwin Quiring, Jonas Möller, Doreen Riepel, Thorsten Holz, Konrad Rieck
Adversarial Learning Text Review Benign Input Reviewer Assignment

October 21, 2022

Evolution of Neural Tangent Kernels under Benign and Adversarial Training
Noel Loo, Ramin Hasani, Alexander Amini, Daniela Rus
Adversarial Attack Adversarial Training Adversarial Robustness Neural Tangent Kernel Specie Evolution Adversarial Vulnerability Benign Input

October 18, 2022

A Novel Feature Representation for Malware Classification
John Musgrave, Temesguen Messay-Kebede, David Kapp, Anca Ralescu
Malware Classification Malicious Sample Benign Input High Resolution Feature

August 5, 2022

FBI: Fingerprinting models with Benign Inputs
Thibault Maho, Teddy Furon, Erwan Le Merrer
Detection Task Benign Input

July 14, 2022

Benign, Tempered, or Catastrophic: A Taxonomy of Overfitting
Neil Mallinar, James B. Simon, Amirhesam Abedsoltan, Parthe Pandit, Mikhail Belkin, Preetum Nakkiran
Training Data Comprehensive Taxonomy Neural Tangent Kernel Model Overfitting Benign Overfitting Benign Input Laplacian Kernel Exponential Measure

February 18, 2022

Debiasing Backdoor Attack: A Benign Application of Backdoor Attack in Eliminating Data Bias
Shangxi Wu, Qiuyang He, Yi Zhang, Jitao Sang
Adversarial Attack Backdoor Attack Data Debiasing Benign Input