1. Memory capacity of recurrent neural networks with matrix representation, with A. Sharma & R. Chandra. Neurocomputing, Volume 560, December 2023.

    We study Fisher information capacity of recurrent networks with matrix representations and compare it with the known results for usual vector representations. We also introduce a new differentiable neural structure with matrix representations throughout and compare it with the traditional neural turing machine on some algorithmic tasks.

  2. C-triviality of manifolds of low dimensions, with S. Sharma. Preprint.

    We classify those low dimensional manifolds upto homology which satisfy the property of $C$-triviality; every complex vector bundle over such a space has trivial Chern classes. Our main tool is the order of differentials of Atiyah-Hirzebruch spectral sequence.