Ayda Sultan

My research interests focus on model interpretability and evaluation, with a particular goal of understanding how models learn, store, and represent concepts. Ultimately, I seek to explore ways to inject and modify concepts within neural networks, advancing our ability to refine and control their learning processes. The big goal is to decode neural network computations neuron by neuron, circuit by circuit, layer by layer.

I was previously a research assistant at KAUST where I worked on concept based interpretability research. Before that, I spent sometime at the University of Michigan working on hand and held-object segmentation. I graduated from Addis Ababa University with a Bachelor’s degree in Software Engineering.