Andrew Luo
@andrew-luo.bsky.social
Assistant Prof at HKU IDS & Psychology, exploring Neuroscience and Machine Learning | Prev: 24' PhD @CMU Neural Computation & Machine Learning | 19' SB @MIT Computer Science | https://andrewluo.net
Forgot to include registers in your ViT? We show that you can introduce registers into your model without *any* labeled data, and significantly improve the quality of dense features.
arxiv.org/abs/2505.21501
"Vision Transformers with Self-Distilled Registers" #AI #ML #ComputerVision #DeepLearning
arxiv.org/abs/2505.21501
"Vision Transformers with Self-Distilled Registers" #AI #ML #ComputerVision #DeepLearning
June 16, 2025 at 3:52 PM
Forgot to include registers in your ViT? We show that you can introduce registers into your model without *any* labeled data, and significantly improve the quality of dense features.
arxiv.org/abs/2505.21501
"Vision Transformers with Self-Distilled Registers" #AI #ML #ComputerVision #DeepLearning
arxiv.org/abs/2505.21501
"Vision Transformers with Self-Distilled Registers" #AI #ML #ComputerVision #DeepLearning
Check out our work on "Meta-Learning an In-Context Transformer Model of Human Higher Visual Cortex" arxiv.org/abs/2505.15813
We design a model which can estimate the voxelwise response in human higher visual cortex using **novel subjects/scanners/imaging-protocols** without any finetuning.
We design a model which can estimate the voxelwise response in human higher visual cortex using **novel subjects/scanners/imaging-protocols** without any finetuning.
May 22, 2025 at 7:23 AM
Check out our work on "Meta-Learning an In-Context Transformer Model of Human Higher Visual Cortex" arxiv.org/abs/2505.15813
We design a model which can estimate the voxelwise response in human higher visual cortex using **novel subjects/scanners/imaging-protocols** without any finetuning.
We design a model which can estimate the voxelwise response in human higher visual cortex using **novel subjects/scanners/imaging-protocols** without any finetuning.