Support Vector Machines

Efficient Dictionary Learning with Switch Sparse Autoencoders
Switch SAEs deliver a substantial Pareto improvement in the reconstruction vs. sparsity frontier for a given fixed training compute budget
Anish Mudide, Joshua Engels, Eric J. Michaud, Max Tegmark, Christian Schroeder de Witt
PDF Cite
Efficient Dictionary Learning with Switch Sparse Autoencoders
Secret Collusion among AI Agents: Multi-Agent Deception via Steganography
We introduce the setting of secret collusion among AI agents.
Sumeet Motwani, Mikhail Baranchuk, Martin Strohmeier, Vijay Bolina, Philip H.S. Torr, Lewis Hammond, Christian Schroeder de Witt
PDF Cite
Secret Collusion among AI Agents: Multi-Agent Deception via Steganography
Unelicitable Backdoors via Cryptographic Transformer Circuits
We introduce encrypted backdoors which cannot be elicited by polynomial-time weight noising.
Andis Draguns, Andrew Gritsevskiy, Sumeet Ramesh Motwani, Christian Schroeder de Witt
PDF Cite
Unelicitable Backdoors via Cryptographic Transformer Circuits
Safe Screening for Support Vector Machines
We present the first safe removal bound for data points whichdoes not rely on spectral properties of the kernel matrix.
Julian Zimmert, Christian Schroeder de Witt, Giancarlo Kerg, Marius Kloft
PDF Cite
Safe Screening for Support Vector Machines