Ivy (lets-unify.ai)

Mentors: Student: Amitesh Vatsa

The aim of this project is to use IVY as a framework and build that into TensorLy project so that TensorLy would be usable with any of the frameworks supported by IVY. This will allow TensorLy users to switch backends via IVY’s backend handler rather than their own native backend handler. Set of deliverables are as follows: Implementing the desired functions in tensorly/backend/ivy_backend.py which will enable users to now use IVY as a framework in TensorLy. Implementing TensorLy’s abstraction layer (functionalities like Tensor Decomposition, Tensor Regression and Tucker Tensors) by using IVY as the framework. Implementing new functions into Ivy’s Functional API which exists in TensorLy’s API but currently not present in IVY’s Functional API (for example: context, index_update, ndim, sign, logsumexp, Conj). Creating demos that demonstrate how the functionalities can be used while building projects using TensorLy when IVY’s backend handler is being used. Running tests and implementing benchmarks showcasing the difference in runtime performance. Documenting the benchmark evaluations in a notebook for further research. Extending features with suggestions from the community.

Mentors: Student: shireenchand

My project proposal for GSoC in Ivy is to work on building Vision Models using Ivy and creating demos and tutorials to showcase the implementation of these models. These models include PSMNet and MLP-Mixer.

Mentors: Student: Dhruv Rajendra Patel

Multi-backend framework support of GradSLAM in Ivy by implementing missing PyTorch APIs: Brief Implementation Plan: 1.) Developing support of a basic low-level API of GradSLAM (which is independent of other GradSLAM APIs) in Ivy frontend and backend (by implementing missing PyTorch APIs), along with its corresponding ivy tests. E.g., gradslam.geometry 2.) Follow a similar implementation strategy for other independent APIs (gradslam.config) or high-level APIs dependent on gradslam.geometry (Eg: gradslam.datasets, gradslam.structures, gradslam.slam, gradslam.odometry) 3.) Develop test script same as in official gradslam repo for all APIs. 4.) Create a tutorial (a jupyter notebook similar to official gradslam notebooks as here) for running GradSLAM using Ivy. 5.) Lastly, showcase better runtime performance of GradSLAM using JAX as the backend framework. Deliverables: -> Multi-backend framework (PyTorch, NumPy, JAX, Tensorflow) support for GradSLAM