Verbalized Sampling, a training-free prompting strategy to mitigate mode collapse in LLMs by requesting responses with probabilities. Achieves 2-3x diversity improvement while maintaining quality. Model-agnostic framework with CLI/API for creative writing, synthetic data generation, and dialogue simulation.

[NeurIPS 2018] [JSAIT] PacGAN: The power of two samples in generative adversarial networks

[CVPR 2024] "Taming Mode Collapse in Score Distillation for Text-to-3D Generation" by Peihao Wang, Dejia Xu, Zhiwen Fan, Dilin Wang, Sreyas Mohan, Forrest Iandola, Rakesh Ranjan, Yilei Li, Qiang Liu, Zhangyang Wang, Vikas Chandra

Mode collapse example of GANs in 2D (PyTorch).

This repository summarizes techniques for KL divergence vanishing problem.

Tensorflow implementation of ICLR2019 paper "Exemplar Guided Unsupervised Image-to-Image Translation with Semantic Consistency"

Multilayer Perceptron GAN, and two Convolutional Neural Network GANs for MNIST and CIFAR.

Dynamic Context Evolution (DCE): Scalable synthetic data generation from a single LLM without cross-batch mode collapse

This 'Generative Adversarial Network' project was implemented in grad course CSE-676 : Deep Learning [Fall 2019 @UB_SUNY] Course Instructor : Sargur N. Srihari(https://cedar.buffalo.edu/~srihari/)

Master's thesis: Adaptive Hybrid Recommendation System with context-aware attention fusion (mode collapse discovery)

Lessons for training cGANs with embedding - housekeeping

Welcome to GAN Studio, a sophisticated, high-performance framework designed for the exploration, training, and deployment of Generative Adversarial Networks (GANs).

Summer 2019 undergraduate research exploring interconnection of Generative Adversarial Networks

Deterministic symbolic algebra prevents mode collapse in LLM synthetic training data. 705 braids, 1410 outputs, 79.5% greater semantic variance. The math is real.

A side-by-side PyTorch comparison of Deep Convolutional GAN (DCGAN) and Wasserstein GAN with Gradient Penalty (WGAN-GP) on the task of anime face generation, demonstrating how WGAN-GP addresses the mode collapse problem inherent in standard GAN training.