Example Topics

LLMs

• Model merging*
• Multi-agent LLMs*
• Large Multimodal Models (LMMs)*
• Retrieval-Augmented Generation (RAG)*
• Scientific discovery with LLMs*
• Confidence estimation / hallucination detection*
• Detecting LLM-generated output*
• Can LLMs know what they know?*

Supervised learning

• Kolmogorov-Arnold networks*
• Reservoir/Liquid State machines*
• Physics-based neural networks*
• Regularization techniques*
• Transformers for visual tasks*
• Transformers for tabular data*
• Bias, fairness, trustworthiness of trained neural networks*
• Transformers for natural language
• Diffusion models (for generating images etc.)
• Foundational language models (e.g. GPT-4)
• Foundational visual models (e.g. DALL-E)
• Foundational video models (e.g. Sora)
• Generative Adversarial Networks
• Neural Turing Machines / Differentiable Neural Computers
• Hyperparameter optimization with gradient descent
• Neural Architecture Search (NAS) with gradient descent
• Simplification of deep networks through distillation
• Second-order methods
• ResNet-type architectures
• Variational autoencoders
• LSTMs for sequence processing
• Image recognition with convolutional networks

Reinforcement learning

• Deep dive to Actor-Critic, DQN methods*
• Policy-search methods (e.g. REINFORCE, natural policy gradients)*
• Neural Architecture Search (NAS) with Reinforcement Learning
• Lifelong reinforcement learning

Evolutionary computation

• Using LLMs as evolutionary operators*
• Evolving LLMs*
• NAS for neuromorphic hardware implementations*
• Optimizing network design to maximize fairness*
• Programming quantum computers with evolution*
• Evolutionary optimization of quantum computers*
• Evolution of language*
• Neutrality, genetic drift, and weak selection*
• Neural architecture search (NAS) with evolution
• Hyperparameter optimization with evolution
• Evolution with uncertain evaluations
• Optimizing aphasia treatments
• Lifelong evolution

Computational Neuroscience

• Networks with spiking neural models*
• Modeling fMRI, DTI, EEG, MEG experiments*
• Using Neuroscience methods to understand LLMs*
• Lifelong learning
• Avoiding catastrophic forgetting

Cognitive Science

• Theory of mind in neural net agents*
• Using methods from Psychology and Social Science to understand LLMs*
• Modeling cognitive disorders
• Modeling aphasia, dyslexia
• Modeling consciousness

Neural network hardware

• Neuromorphic (spiking) designs*
• Quantum neural networks*
• Hardware accelerators for deep learning
• Optical, molecular, atomic neural networks

Applications

• Stock market prediction*
• Optimizing medical treatments*
• Automated driving*
• Natural language understanding
• Document processing
• Speech understanding
• Robotic control
• Medical image understanding

Theory

• Regularization*
• Evolutionary optimization*
• Advantages of overparameterization*
• Super-Turing computing with neural networks
• Effective vector embeddings