SajayR
@sajayr.bsky.social
CV/NLP Undergrad Researcher
I post about cool papers in CV/NLP or about my own fun experiments :D
I post about cool papers in CV/NLP or about my own fun experiments :D
Releasing an *extremely* early research checkpoint of Triad - learning to map precise, localized connections between vision, audio, and text features in a shared space with semantic concepts embedded in each patch/segment/span itself instead of global understanding.
March 2, 2025 at 12:21 PM
Releasing an *extremely* early research checkpoint of Triad - learning to map precise, localized connections between vision, audio, and text features in a shared space with semantic concepts embedded in each patch/segment/span itself instead of global understanding.
I mean-this might be a done thing but like learned ViT patch truncation *might* work? I am just playing around for now but I have some interesting ideas I wanna try
January 17, 2025 at 4:21 PM
I mean-this might be a done thing but like learned ViT patch truncation *might* work? I am just playing around for now but I have some interesting ideas I wanna try
Reposted by SajayR
🚀 Excited to present NeuralSolver at NeurIPS 2024!
Humans can easily learn algorithms to solve much more complex tasks, from simple tasks. Machine learning methods usually fail to do this.
Can we build models that have similar algorithmic extrapolation abilities?
Thread 🧵:
Humans can easily learn algorithms to solve much more complex tasks, from simple tasks. Machine learning methods usually fail to do this.
Can we build models that have similar algorithmic extrapolation abilities?
Thread 🧵:
November 26, 2024 at 2:22 PM
🚀 Excited to present NeuralSolver at NeurIPS 2024!
Humans can easily learn algorithms to solve much more complex tasks, from simple tasks. Machine learning methods usually fail to do this.
Can we build models that have similar algorithmic extrapolation abilities?
Thread 🧵:
Humans can easily learn algorithms to solve much more complex tasks, from simple tasks. Machine learning methods usually fail to do this.
Can we build models that have similar algorithmic extrapolation abilities?
Thread 🧵: