Abstract:Original data usually lies on a set of disconnected manifolds rather than a smooth connected manifold. This causes the problem of mode collapse in the training of vanilla Generative Adversarial Network (GAN). There are many existing GAN variants that attempt to address this problem, but they result in limitations. The existing variants either produce simulated instances with low quality or generate identical simulated instances. In this study, we propose a new approach to training GAN utilizing multiple generators, a classifier and a discriminator to address mode collapse. The classifier outputs the statistical probabilities of generated data belonging to a specific category. These probabilities implicitly reflect which manifolds are captured by generators, and the correlation between generators is quantified by mutual information. Our idea views the mutual information values as a constraint to guide generators in learning different manifolds. Specifically, we traverse the generators, calculating the mutual information between each generator and the others. The calculated values are integrated into the generator loss to form a new generator loss and to update the corresponding generator's parameters, using back-propagation. We minimize the mutual information to reduce the correlation between generators while also minimizing the generator loss. This ensures generators capture different manifolds while updating their parameters. A new minimax formula is established to train our approach in a similar spirit to vanilla GAN. We term our approach Mutual Information Multi-generator GAN (MIM-GAN). We conduct extensive experiments utilizing the MNIST, CIFAR10 and CelebA datasets to demonstrate the significant performance improvement of MIM-GAN in both achieving the highest Inception Scores and producing diverse generated data at different resolutions.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is to understand how the excitatory synapses of different subpopulations of midbrain dopamine (DA) neurons are specifically regulated when they are stimulated by rewards or aversions. Specifically, the researchers want to explore whether this regulation is related to the brain regions to which these DA neurons project. Through this research, they hope to reveal that the mesocorticolimbic system may consist of three anatomically distinct circuits, each of which responds to different aspects of motivation - related stimuli, including rewards, aversions, and salience. ### Main findings: 1. **Different DA neuron subpopulations**: The study found that, depending on the target regions to which DA neurons project, their basic synaptic properties also vary. 2. **Effects of reward stimuli**: - The excitatory synapses of DA neurons projecting to the medial shell of the nucleus accumbens (NAc) were significantly enhanced after cocaine administration. - The excitatory synapses of DA neurons projecting to the medial prefrontal cortex (mPFC) did not change significantly after cocaine administration. - The excitatory synapses of DA neurons projecting to the lateral shell of the NAc were also enhanced after cocaine administration, but not as obviously as those of neurons projecting to the medial shell of the NAc. 3. **Effects of aversive stimuli**: - The excitatory synapses of DA neurons projecting to the mPFC were significantly enhanced after aversive - stimulus administration. - The excitatory synapses of DA neurons projecting to the medial shell of the NAc did not change significantly after aversive - stimulus administration. - The excitatory synapses of DA neurons projecting to the lateral shell of the NAc were also enhanced after aversive - stimulus administration. 4. **Long - term effects**: The enhancement effect of the excitatory synapses of DA neurons projecting to the medial shell of the NAc after cocaine administration can last for 21 days, while neurons in other subpopulations do not have such long - term effects. ### Conclusion: These results suggest that the dopamine neurons in the mesocorticolimbic system may be composed of multiple parallel circuits, each of which makes specific responses to different aspects of motivation - related stimuli (such as rewards, aversions, and salience). This provides a new perspective for further understanding the role of the dopamine system in motivation control.

Multi-generator GAN learning disconnected manifolds with mutual information

X-Gacmn: An X-Shaped Generative Adversarial Cross-Modal Network With Hypersphere Embedding

Statistics Enhancement Generative Adversarial Networks for Diverse Conditional Image Synthesis

Dual Distribution Matching GAN

Tackling mode collapse in multi-generator GANs with orthogonal vectors

Combating Mode Collapse in GANs via Manifold Entropy Estimation

Dual Discriminator Weighted Mixture Generative Adversarial Network for image generation

Shared Loss between Generators of GANs

MIM-Based GAN: Information Metric to Amplify Small Probability Events Importance in Generative Adversarial Networks.

Generative Multi-Adversarial Networks

Study of Prevention of Mode Collapse in Generative Adversarial Network (GAN)

InfoMax-GAN: Improved Adversarial Image Generation via Information Maximization and Contrastive Learning

Improving mode exploring capability of generative adversarial nets by self-organizing map

STM-GAN: Sequentially Trained Multiple Generators for Mitigating Mode Collapse

MMD GAN: Towards Deeper Understanding of Moment Matching Network

Mutual Information Compensation for High-Fidelity Image Generation With Limited Data

Generative Adversarial Mapping Networks

Combating Mode Collapse via Offline Manifold Entropy Estimation

MRD‐GAN: Multi‐representation discrimination GAN for enhancing the diversity of the generated data

Towards the Gradient Vanishing, Divergence Mismatching and Mode Collapse of Generative Adversarial Nets