About

Hi, I’m Akash.

🌐 github.com/akashpalrecha

E-mailTwitterGithubLinkedIn+919879990466

Resume.pdf


👨🏻‍🎓 Background


I work as an AI Researcher in a Space-tech startup (Pixxel) that is building a constellation of earth-imaging small satellites 🛰 to provide an entirely new kind of dataset of the earth that today’s satellites aren’t capable of.

I am currently pursuing an MSc. Mathematics degree in Birla Institute of Technology and Science, Pilani as a pre-final year student. My best work primarily centres around making AI algorithms for solving computer vision problems such as classification and segmentation. My research interests include transfer learning, optimizers, interpretability and everything that lowers the number of resources required to do great, world-class work and makes deep learning more accessible. I like reading new research papers every week and look forward to getting into NLP.

I like tackling big, high impact problems. One such problem I believe in is making AI accessible. I am greatly influenced by Jeremy Howard’s (Co-Founder, Fast.ai) approach to AI and my belief in making AI accessible stems from Fast.ai’s core philisophy of making neural nets uncool again.


In my free time, I like helping out college students to get into deep learning. I also play drums with my college band in BITS Pilani’s Music Club. We make new songs all the time and have a great time jamming together every day!


Experience



A day in my life
- College classes.
- Finding new, impactful Research papers to send to my Kindle (Twitter’s AI Community helps).
- Work in Pixxel AI, and on personal research projects.
- Binge-surfing forums.fast.ai.
- Go Run!
- Jam/Compose with my band post dinner (I play drums!)
- Learn, learn, learn!


👨🏻‍💻Working On / Learning


Research

  1. A modification to Batchnorm layers that doesn’t require mean or variance parameters and gives comparable or better performance. (personal research)
  2. A general method to classify images using deep models by taking assitance from query images.

Learning

  1. FastAI V2
  2. Part 2 (lecture 12) of FastAI’s Deep Learning Course
  3. (Week 8) CS50’s Web Programming with Python and Javascript

Blogs/Tutorials:

More blogs coming soon


💻 Projects / Small Code Sprints


Training Deep Models:

From Scratch (Utilities, Practice):

I believe that being able to build things for yourself when you can’t find open-source code (libraries) is an important skill. This is a small subset of the things that I’ve needed to build over time:

Development


Skills: