Avatar

Soorya Gopalakrishnan

About me

I recently received my PhD in Electrical and Computer Engineering from UC Santa Barbara, where I was advised by Prof. Upamanyu Madhow. My research is broadly on signal modeling for robust neural networks, with focus in the following areas:

  • Robustness to adversarial examples
  • Robust radio frequency (RF) signatures

Interests

  • Robust machine learning
  • Signal processing
  • Compressive sensing

Education

  • PhD in Electrical and Computer Engineering, 2020

    University of California, Santa Barbara

  • B. Tech and M. Tech in Electrical Engineering, 2014

    Indian Institute of Technology Madras

Research

Robustness to Adversarial Examples

Combating adversarial perturbations by exploiting sparsity of natural data.
PDF Code

Robust RF Signatures

Learning signatures to distinguish between wireless devices sending exactly the same message.
Slides

Publications

Soorya Gopalakrishnan, Zhinus Marzi, Upamanyu Madhow, Ramtin Pedarsani (2020). Robust Adversarial Learning via Sparsifying Front Ends. arXiv:1810.10625.

PDF

Metehan Cekic, Soorya Gopalakrishnan, Upamanyu Madhow (2020). Robust Wireless Fingerprinting: Generalizing Across Space and Time. arXiv:2002.10791.

PDF

Soorya Gopalakrishnan, Metehan Cekic, Upamanyu Madhow (2019). Robust Wireless Fingerprinting via Complex-Valued Neural Networks. IEEE Global Communications Conference (Globecom).

PDF

Zhinus Marzi, Soorya Gopalakrishnan, Upamanyu Madhow, Ramtin Pedarsani (2018). Sparsity-based Defense Against Adversarial Attacks on Linear Classifiers. IEEE International Symposium on Information Theory (ISIT).

PDF

Soorya Gopalakrishnan, Zhinus Marzi, Upamanyu Madhow, Ramtin Pedarsani (2018). Combating Adversarial Attacks Using Sparse Representations. International Conference on Learning Representations (ICLR) Workshop Track.

PDF

Soorya Gopalakrishnan, Tiffany Moy, Upamanyu Madhow, Naveen Verma (2017). Compressive Information Acquisition with Hardware Impairments and Constraints: A Case Study. IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP).

PDF