Human Learning Journal: Working from home

Original article was published by Valentin Kindschi on Deep Learning on Medium


Human Learning Journal: Working from home

This weekly journal reports my up and downs as a machine learning and robotics intern at the impactIA Foundation. Try to have fun reading it, but know that I have little fun writing it. You have been warned.

This week was the second week in a row I worked from home every day. It was the first time I stayed at home that long to work and I wanted to share my thoughts about it. Similarly to last week, I focused on tuning deep neural networks for image classification, but without the help of AutoKeras. With my colleague Léo, we experimented with the layers size and depth, played with the learning rate and tried various image augmentations, aiming for a lower loss and a higher accuracy.

As I said in my last blog, the main challenge in our project is to work with real world data, and not cat vs dogs images. To tune our deep neural network parameters, we conducted a lot of experiments, to observe their impacts on the results in our specific case. It was really important to document each of these experiment, write down the hyper parameters, the performance of the resulting network and the progress we were making.

Since I arrived at impactIA, I have been discovering different projects such as Dai, the aixlr8 program and our AI toolkit. Every time I was describing to my colleagues what I was doing and my ideas about them, but honestly I find it hard to communicate efficiently and precisely when we meet physically only once a week. Therefore, this week I changed a little bit my way of working. I decided to write down all the tests I was making and my conclusions about them in a shared file. I gave a link to the file to all my colleagues that I was working with so that they could see what I had been up to.

This approach had two big advantages for me. First, writing down problems, thoughts and solutions really helped me to have a better understanding of what I was doing. I have the feeling that the tests I was making and the decision I was taking were more thorough. Second, sharing all the details of my experiments with the team allowed them to give me quick and insightful feedback, which led to a more structured teamwork and faster progress.