|Fast and Memory Efficient Linear Transformation with Structured Matrices|
- Circulant Matrices
- Kronecker Factored Matrices
- Hadamard-Diagonal Matrices
- Applications in deep learning, binary embedding, kernel approximation, product quantization
|Learning with Label Proportions (LLP)|
- Can we predict the individual labels given only some label statistics on groups?
- Broad applications (and privacy concerns) in political science, marketing, healthcare etc.
- Theoretical analysis to understand when and why LLP is possible.
- Solving challenging problems in computer vision.
|Attribute-based Visual Recognition
- Framework and methods for designing discriminative visual attributes for recognition
- Using attributes for large-scale image retrieval