Steffen Schotthöfer and Emanuele Zangrando from our lab are attending NeurIPS Conference this week in person and will present our work on lowrank training and pruning of neural networks. In this work we developed a framework to perform stable and efficient training on low-rank manifolds, resulting in an order of magnitude less memory cost and training time! Tested successfully on Imagenet1K, transformers and several other benchmarks.

If you are there too, swing by our poster session in HallJ#604 on Wed 30 Nov 9:30 am PST