Sobolev Space Regularised Pre Density Models

Authors: Mark Kozdoba, Binyamin Perets, Shie Mannor

ICML 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We evaluate the resulting method on the comprehensive recent anomaly detection benchmark suite, ADBench, and find that it ranks second best, among more than 15 algorithms.
Researcher Affiliation Collaboration 1Technion Israel Institute of Technology, Haifa, Israel 2NVIDIA Research.
Pseudocode Yes Algorithm 2 Density Estimation Procedure with Hypeparameter Tuning, Algorithm 3 K( ) R2 N : Sampling the Multidimensional SDO Kernel, Algorithm 4 Optimal_alpha( ) RN : Calculating the Optimal Alphas, Algorithm 5 F 2( ) RNY : Density over coordinates Y given observations X., Algorithm 6 FD( ) R : Fisher Divergence with Hessian Trace Approximation.
Open Source Code Yes All code necessary to replicate the results presented in this paper, as well as the implementation needed to run SOSREP with the kernels described herein, is publicly available. You can access and download the code from our Git Hub repository at (https://github.com/bp6725/SOSREP).
Open Datasets Yes ADbench (Han et al., 2022) is a recent Anomaly Detection (AD) benchmark evaluating 15 state-of-the-art algorithms on 47 tabular datasets.
Dataset Splits No The seed determines the train-test split of the ADbench data, as well as the randomness of the algorithms.
Hardware Specification Yes As for computational cost, the entire set of 47 datasets was processed in 186 minutes using a single 3090RTX GPU and one CPU, averaging about 4 minutes per dataset.
Software Dependencies No automatic differentiation, offered in frameworks such as Py Torch.
Experiment Setup Yes The bandwidth parameter a for SOSREP was selected based on the best LLK from a verification set using a grid search over a grid of 10 values. Algorithm 4 Optimal_alpha( ) RN : Calculating the Optimal Alphas Require: X, lr, niters