                     XXXXXXX
                     XXXXXXX
                    XXXXXXXX
                    XXXXXXXX
                    XX      
           XXX       X      
          XX X       X      
         XX  XX      XX     
         X    XXXXXXX X     
        XX          XXX     
        x                   
        x                   
xxx     X                   
XXX    XX                   
XXX  XXX                    
  XX X                      
   XXX

 XXX XXX    XXX XXX        XXX      XXX XXX              
XXX   XXX        XXX     XXX XXX    XXX   XXX            
XXX   XXX       XXX     XXX   XXX   XXX   XXX            
XXX   XXX    XXX XXX     XXX XXX    XXX   XXX            
XXX   XXX     XXX       XXX   XXX   XXX   XXX            
XXX   XXX    XXX        XXX   XXX   XXX   XXX            
 XXX XXX      XXX XXX   XXX   XXX   XXX   XXX            

 XXX XXX       XXX      XXX   XXX    XXX XXX    XXX XXX  
XXX   XXX    XXX XXX    XXX   XXX      XXX           XXX 
XXX   XXX   XXX   XXX    XXX XXX       XXX          XXX  
 XXX XXX     XXX XXX       XXX         XXX       XXX XXX 
XXX   XXX   XXX   XXX      XXX         XXX        XXX    
XXX   XXX   XXX   XXX      XXX         XXX       XXX     
 XXX XXX    XXX   XXX      XXX       XXX XXX      XXX XXX

Ozan Bayiz

I'm a fourth-year Regents' and Chancellor's Scholar at UC Berkeley studying Computer Science. I'm interested in exploring efficient video model adaptation for multimodal conditioning, and applying these techniques to support AI-enhanced learning.

These days I spend most of my time in BAIR or TAing for CS 189/289A: Intro. To Machine Learning.

I'm extremely grateful for Vongani Maluleke, who has been an amazing mentor to me as I navigate my odd research journey.

Figure 1B: Ozan and Caramel (Peaceful).

Featured Projects

all projects

ozanbayiz.net
This website! built with React, Next.js, Tailwind CSS, shadcn/ui, and lots and lots and lots of love.
Investigating Demographic Attribute Representation in Vision Encoders
Probed Florence‑2’s DaViT vision encoder with linear classifiers and trained patch‑level SAEs to discover interpretable sparse dictionary features (SDFs).