ICLR 2026 Notes

This is an unordered, unstructured list of thoughts that occured during the ICLR 2026.

• For the graduation project, how about using a small local LM (e.g. Gemma 3n or another MOE)?
• When working on AI projects, focus on augmenting humans, not replacing them. Market your project as such.
• Follow your Intellectual Curiosity !!!
• Project Idea: Follow-up work on Inducing Dyslexia in VLMs. Using the VLMs for something other than font suitability prediction for dyslexic individuals.
• You should read the model guided microstimulations paper by EPFL NeuroAI
• Project Idea: Learn more about Waves-as-Memory. Work on implementing a small part related to it. You can also attmpt to apply it to a toy dataset or example.
• Project Idea: Try building a toy example related to Akron paper
• Read the Equivariant GNN paper and try a small related project
• There is a huge field concerned with the generation of 3D worlds and their extraction from images or videos
• Interesting paper from the vector instituite: “Homeostatic Adaptation of Optimal Population Codes Under Metabolic Stress”
• There is an interesting world of molecules generation and drug discovery
• You may tree eInx which is a cool interface over PyTorch, np, Jax, and others for tensor operations.
• The world of temporal graph learning is concerned with ML on graphs that change over time
• Computational Imaging is BEAUTIFUL !!!
• Energy Based Models are being used for constraint satisfaction problems !!!
• There is a world concerned with the generation of graphs

Discovered Research Areas

They are not written in any particular order, nor do they have the same level of granularity.

• Computational Imaging
• Agentic AI and LLMs
• RL
• Neuroscience
  • BCI
  • Spiking Neural Networks
• Graph ML
  • Graph Generation
  • Relational Deep Learning
  • Temporal Graph Learning
• Physics X AI
  • Other forms of memory
  • thermodynamics X AI
• Work on improving architectures of machine learning models (advances in diffusion, EBT, etc.)
• Infographics Generation
• Presentation Generation
• Image Editing
• Vectors (SVG) Generation
• Graphic Design
• Chemical Molecules Generation
• GNN + LLMs | RAG on Graphs
• MOE
• 3D World Generation and Extraction
• World Models
• Geneomics
• Protein Design
• Wearable Health Devices and Biosignals
• Time Series FOundation Models