Welcome
My name is Ignasi Mas, I am a Machine Learning/AI Engineer with a broad interest in solving data issues. My background is built in Computer Vision, although my interest focus (which obviously includes Computer Vision) is wider.
About me
I studied Telecommunication Engineering at ETSETB, UPC. Some day we may talk deeply about my experience studying this career, but for now let’s just say that the content in it is acquiring the knowledge to understand each point in the signal lifecycle.
What’s the particularity about this? Well, our knowledge about signal has increased so much during the last decades, that more and more issues and solutions have flourished, thus becoming Telecommunication Engineering a vastly wide career. That is of course something good but is also more sensible to unrelated limitations. In the end, time is limited, and if you want to cover everything you lose granularity.
During my progress in my career, I felt more and more attracted to the development of intelligent systems. In that context, I demanded more knowledge than what I was getting, but the loss of granularity mentioned above made that impossible. Once I graduated I felt incomplete about that. I missed something. And that something was Machine Learning. Based Andrew NG opened my eyes through its course.
I found the following piece in my career’s puzzle as the Master in Computer Vision from Computer Vision Center. There perhaps I may be able to merge two of my main academic interests, Machine Learning and Computer Vision. Two years later, I didn’t regret that decision. This Master fed me with the seeds to gain further knowledge, and begin learning everything in the wild.
But my adventure still had to deliver another incredible chapter. That was the development of my Master’s thesis. I wanted to focus on something of my special interest, so I researched open and hot topics in Machine Learning. I had many interests so it was hard to choose, but I found in Few-Shot Learning one of my main focuses. There, I learned about Meta-Learning and found one of the potentially needed from Meta-Learning problems in ML: Active Learning.
In this blog, we will have time to talk about Meta-Learning and Active Learning further, but just know that I developed my Master’s thesis about Meta-Active Learning. I.e. using Meta-Learning to solve Active Learning problems. I focused on one concrete Active Learning scenario (again, we will study these scenarios someday in this space). I sweated blood just to replicate the State of the Art approaches. I remember the days trying to handle memory in my PyTorch tensors… (in Meta-Learning, memory management works differently since you keep different gradients for different levels… but no more spoilers). And I almost jumped out of joy the day I had some reasonable results in my proposed solutions.
This chapter did not finish presenting my dissertation for the Master (which I obviously did and finally got), but I presented it to the MDALC Workshop at ICCV 2019, and they accepted it! We published the paper and it is accessible to anyone since then.
Is my academic history finished yet? Of course not! I am actively thinking about a possible PhD that I may do someday. I actually had a couple of opportunities that in the end were not materialized at all, but for sure there will be more. It is not something time sensitive right now, but it would be another way to acquire and deliver more knowledge. Actually, this will not necessarily be my only path, I may find other ways to do so. So new adventures await!
Oh, but Ignasi, you didn’t tell us about your professional trajectory. Yes, indeed. I have plenty of experiences and cool projects along with incredible people that I participated in. But that is probably a story for another space. Or maybe another day…
About this blog
In the current episode, I realized that I spend time browsing the State of the Art in some matters and playing with code, but it is something I have always done by myself. However, wait…. Why not share it with everyone? It is a good trade. You get my knowledge and I get your feedback. So, how can I share it with everyone? Oh, a blog! It is an easy tool where I may focus on the content, instead of the shape. If this grows, I may redefine it later. But at my beginnings, that is the idea.
At the time I am writing this, I still have to do everything in the blog. My idea here is to post in a more or less formal way (I want to make it easy to read for you) posts about different theory ML topics of my interest as well as maybe some practical exercises where I’ll share my reasoning live.
I can’t tell you the frequency at which I will be posting. My idea first is to try to post every two weeks (time enough for researching some problem and preparing the post while I am working because I need to eat, you know?), but I can’t promise anything yet.
In the first weeks, I will post topics I already know about and focus on how to present them to you. This way, I will train myself for the future, in which I will deliver to you new topics.
So just one more thing: let’s have fun together!