For our latest machine learning specialist interview on our blog, we’ve welcomed Ivan Goncharov, a machine learning engineer on the growth team at a unicorn MLOps startup, Weights & Biases.
Tell us about the business you represent, what is their vision & goals?
I work at Weights & Biases, which is an MLOps platform aiming to build the best tools for machine learning practitioners.
Can you share a little bit about yourself and how you got into the field of machine learning?
I got into the field of Machine Learning during my last year at high school after being inspired by the possibility of computer vision helping people with disabilities.
In particular, I teamed up with a friend of mine and we developed (and open-sourced) an Android application capable of helping visually impaired people by reading bus route numbers for them. I eventually made a video explaining the code and the approach on my channel.
Can you tell us how you started creating content for YouTube?
I haven't trained academically in machine learning, so, the only source of knowledge that I myself started learning was found online in the form of videos, blog posts and research papers.
I've always enjoyed tinkering with the video format and really enjoyed the process of helping others understand concepts that I personally found fascinating. And I also - having gone down the self-taught path - understood well enough the struggle that it can be to learn something new. So, I started making my first videos literally explaining how I trained certain neural networks or collected data.
Many of the videos were inspired by the BusNumberApp project I talked about above. Over the years, the quality has gone up but the underlying principle, I would say, is still the same: make something useful and have fun along the way.
What does your day to day responsibilities look like at your organisation?
I'm on the growth team at Weights & Biases, so my main responsibilities are the creation of content around machine learning in the form of videos and blog posts that explain something and, ideally, include the usage of products in them. And also handling the many different ever-changing tasks that come with working for a quickly-growing startup.
What are some misconceptions that you believe the average person has about machine learning?
I can't speak for how many people have this misconception but I would say that, maybe, it's that you need to somehow be a genius to do machine learning, or that it's not for everybody?
I'd say that the field is so large these days - with many different job positions and ways that it can be applied. And vastly different specialists are required beyond just the machine learning researchers and theorists. So, it's a really big field and if you are fascinated by it, then you can work in it doing many different things.
What advice would you give to someone wishing to start their career in machine learning?
I would say make stuff. The best way to learn, in my opinion, is by doing. And the magical thing about machine learning (and maybe computer programming in general) is that you don't need any special equipment besides your laptop to start making amazing things.
What about the computing for machine learning, you may ask? Well, services like Google Colab and Kaggle provide free GPU instances in the cloud that are more than sufficient for any beginner projects you may take up. Just think about something cool that you'd like to work on (for example a former student of mine worked on sign language recognition, which inspired this video of mine), look up how to make it online, and make it.
What is your experience with using AI-backed data analysis?
A couple of months ago I worked on a project analysing stock market news headlines with FinBERT (a transformer network) and used Weights & Biases to analyze the model's (FinBERT's) predictions.
It was interesting because for the financial domain normal natural language processing models don't really generalise well because of all the specific jargon, so a specially fine-tuned model on the financial corpus of data is really needed to boost the results.
Are there any tools that you swear by?
Jupyter Notebooks are really great for tinkering with machine learning or data cleaning applications, also I have had great experiences with TensorFlow and Keras deep learning libraries. And also, Weights & Biases is a really cool MLOps platform for tracking experiments.
Are there any books, blogs, or any other resources that you highly recommend on the subject of AI?
Something that has helped me a lot in my journey and that I recommend to almost everyone is 3Blue1Brown's videos on the intuitions of artificial neural networks and the free book they're based on neuralnetworksanddeeplearning.com. Also, for understanding convolutional neural networks the famous CS231n course is really good.