HiSS

This repository contains code for the paper Slithering through Gaps: Capturing Discrete Isolated Modes via Logistic Bridging, accepted in International Conference on Artificial Intelligence and Statistics (AISTATS), 2026.

@article{mohanty2026hiss,
  title={Slithering through Gaps: Capturing Discrete Isolated Modes via
Logistic Bridging},
  author={Mohanty, Pinaki and Zhang, Ruqi},
  journal={International Conference on Artificial Intelligence and Statistics},
  year={2026}
}

Introduction

We propose Hiss, a discrete sampler for sampling from landscapes with disconnected modes.

Dependencies

• PyTorch 1.9.1
• torchvision 0.10.1

Usage

Sampling From 4D Joint Bernoulli

Enter Directory

./Bernoulli

Then run

python bernoulli_sample.py --sampler=<SAMPLER>

Sampling From Ising Models

Please run

python ising_sample.py --sampler=<SAMPLER>

Travelling Salesman Problem

Enter Directory

./TSP

Then run

python TSP.py --sampler=<SAMPLER> 

Binary Bayesian Neural Networks

Enter Directory

./BinaryBNN

Then run

python bayesian_nn.py --sampler=<SAMPLER> --dataset=<DATASET>

References

• This repo is built upon the DLP repo