Hello everyone,

My name is Spiderman, and I'm currently pursuing my Master's degree in Computer Science at IIT Guwahati, one of India's leading institutes. My specialization is in Machine Learning, Deep Learning, and Neural Networks. I want to share my experiences and insights with all of you who are interested in this exciting field.

Areas of Expertise

In machine learning, I am comfortable with everything starting from the basics, like supervised and unsupervised learning, moving towards deep learning. Currently, my area of expertise is Natural Language Processing (NLP). I'm also into Generative AI (Gen AI), including Large Language Models (LLMs) and related technologies. I've worked on some projects in computer vision as well. So, I've covered everything from basic machine learning to advanced Generative AI courses.

Understanding NLP and Key Topics

Advanced topics in NLP start from transformers. We all know the "Attention Is All You Need" paper published by Google in 2017, which introduced the transformer model. It starts from the transformer encoder-decoder model and extends to areas like entity linking and more advanced topics like Retrieval-Augmented Generation (RAG) based LLMs.

• Everything in NLP now is based on transformers—even models like ChatGPT work on that architecture.

• I'm also working with Hugging Face, where there are many pre-trained models available.

• We can explore how to fine-tune these models and use them in whatever area we want to.

• By exploring these topics and models, we can build something of our own, which can be unique and could even be converted into a startup.

Exploring Generative AI

Generative AI is wholly based on the transformer model. So, what exactly is Generative AI?

• Traditional machine learning models were trained on some data and would generate outputs based on that data.

• But now, with Generative AI, the models can think on their own and generate something entirely new.

• For example, we have ChatGPT. Given some input, it generates answers that are very creative and original.

• The basic core model here is the encoder-decoder model, and that's where it all starts.

• If you know the basics, you can build something great on those foundational blocks.

• Generative AI is a very promising field, and everything will be moving towards it in the future.

Real-Life Projects in NLP

To explore the Generative AI field, I think one should start with Hugging Face. It's an open-source library with many pre-trained models available in every field—be it images, audio, or text.

• You can take any model and fine-tune it according to your needs.

• Let's say we want some image generators for a specific task.

  • For example, if we want to generate images of bottles only, we have models that already generate new images, but we can fine-tune them.

• You just have to tell the model that you want bottle images, and it can create anything—the level of creativity can be beyond what we can think of, and it can generate it.

• This technology can, in some ways, replace creative artists because generative techniques can produce new designs and concepts.

• There are many things that can be done using Generative AI.

• So, I would suggest starting with Hugging Face, exploring it and the models, and then getting started.

Tools and Topics for Students

The basics you need include Python and various frameworks based on Python like PyTorch. You can get started from there. You should know basic machine learning and deep learning, how to fine-tune those models, and how to use them.

Later on, there are some theoretical concepts you should know. It depends on which area you want to work in:

• Text and Language: Focus on NLP.

• Images: Learn about computer vision and discriminative models.

• Speech: Explore models and techniques specific to audio processing.

It depends entirely on which area you want to work in, but the basic stuff remains the same: machine learning, deep learning, transformer architectures, and Python programming.

My Master's Projects and Thesis

I have always been interested in these fields, so I explored various areas during my Master's:

1. Speech Processing:

   • I took a course in speech processing where I worked on projects related to speech, such as automatic speech recognition.

   • For example, detecting what digit or word someone says.

2. Image Processing:

   • I took an image processing course in which we made projects related to image processing.

   • We created an automated entry system using generative models to detect which ID it is—is it a college ID, Aadhaar card, or something else.

3. Natural Language Processing:

   • Now, I'm working on my thesis, which is based on NLP.

   • I took courses related to NLP and made projects aiming to publish papers in this field.

   • My thesis focuses mostly on entity linking.

Research Papers in the Field

Research papers in this field are getting published on topics like entity linking and RAG techniques. RAG stands for Retrieve and Generate. It's an advanced technique that is not widely used currently.

• For example, ChatGPT knows some generalized data, but if you want to give it specific information, like industry knowledge or domain-specific content, we use these techniques to retrieve the data and generate accurate answers.

• By using RAG, we can enhance models to provide more precise and relevant outputs based on specialized data.

Opportunities and Experience in the Field

My experience in this field has been great. Every day, things change, and new advancements are coming. Everyone knows how AI is growing and where it's going.

• All the leading industries, starting from Google, Meta, and OpenAI, are working in this field of Generative AI.

• I think it has a very bright future, and everything will be based on these technologies in the coming years.

Keeping Up with Rapid Changes

As I mentioned, the field is changing at a very high speed. You might wonder if what we're learning today will be outdated tomorrow. Even if it's changing every day, the basics remain the same—the core concepts of machine learning and deep learning are consistent. We can use recent advancements, fine-tune existing models, and make something better according to our needs.

You have to accept that you might be working on something, and it has already been discovered before you publish it. But you have to adjust, build on top of it, and continue innovating.

It's a competitive world, but it's fun, really exciting, and it's the future.

Advice for Aspiring Students

There are many bachelor students who are aspiring to enter this field. In their bachelor's programs, they might not have courses related to ML and AI. Even so, they can prepare themselves for a Master's in NLP and Generative AI.

I recommend starting to build your basics. You can do this using some very famous courses on Coursera, like those by Andrew Ng—everyone might know him. His courses on machine learning and deep learning are really good. Even I started from there.

There is also Andrej Karpathy, a Stanford professor, who has provided some really good insights in the field of transformers and related topics. Additionally, we have an Indian professor named Mitesh Khapra. His lectures on deep learning are great, even in the field of Generative AI. There are many resources available online—YouTube, online courses, and tutorials.

Qualities Needed for AI and ML Engineers

Students who are aspiring to become AI and ML engineers might wonder what capabilities they should have, or how they can evaluate themselves during their bachelor's studies.

• You don't need to be super smart to be an AI engineer. Even I consider myself a normal, average person.

• The key is to be clear with the basic concepts, and consistency is crucial.

• Having an interest and passion for new technology and innovations also helps a lot.

Final Thoughts

Thank you for giving me the opportunity to share my insights and experiences while doing my Master's at IIT Guwahati. I hope my journey helps others in their path toward becoming AI and ML engineers.

Remember, the field is evolving rapidly, but with strong basics and continuous learning, you can adapt and excel.

Wishing you all the best in your endeavors!

Warm regards,

Spiderman