[~]$

whoami

Hi, I am an AI Software Engineer @ Qualcomm. I’m passionate about building intelligent systems at the edge. My expertise lies in Embedded Systems, On-Device AI, Reinforcement Learning, and Machine Learning. I’m always eager to learn and collaborate on innovative projects. Read more..

You can reach out to me at mynkpl1998@gmail.com.

Tokenization

What is tokenization ? Machine learning models operate on numerical representations of input data. Language models, such as LLMs, work with text input. Tokenization is the process of converting the text into smaller chunks called tokens, which are then assigned numerical representations. The collection of unique tokens forms what we call a vocabulary. It is essentially a dictionary that maps each token to a unique integer. The size of the vocabulary is determined by the number of unique tokens in the corpus....

April 19, 2024

Bit Manipulation

Bitwise Operators The following are the bitwise operators which are available in C/C++ programming language. Operator Meaning & Bitwise AND | Bitwise OR ^ Bitwise XOR ~ Bitwise complement « Shift left » Shift right XOR The main property of XOR is that it keep the bit same if both operands are same, else it flips the bit. A B Result 0 0 0 0 1 1 1 0 1 1 1 0 Few Properties of XOR Property Result A ^ 0 A A ^ 1 ~A A ^ A 0 A ^ A ^ A A A ^ B ^ A B A ^ B ^ B A Few properties of Shift operators....

April 8, 2024

Dissecting Policy Optimization

I assume the reader is familiar with the basics of Reinforcement learning and has a basic understanding of statistics and a bit of calculus. One should be comfortable with manipulating value functions, policy, and bellman equations. The main idea of writing this blog post is to summarize and extend the understanding of reinforcement learning methods that directly optimizes policy. More or less, this blog post is a summary for me to revisit the concepts and various tricks that are helpful while dealing with Policy-based optimization....

June 27, 2020