What progress did I make?
Wow, can’t believe how quickly time has flown by. With regards to my MLSA, there has been both positive developments and negative occurrences, I’ve learned how to be better from both. One of the major things I’ve realised is how critical it is that I approach this endeavour with maximum integrity! Part of this means being really genuine and honest with the struggles I’ve been facing so that if others ever feel like giving up or they can’t make it or they aren’t good enough to be in this field, they can look at my example and know that I’ve felt just like that too. It’s totally human and it’s ok, what matters is that you don’t given up and keep going, doing the best you can with where you are. As long as you keep being tenacious, reflecting on mistakes and growing, then any vision can be created!
✅ Kaggle learn: Intro to game AI and RL
I explored this mini-course on Kaggle which was quite helpful in showing me how you could approach designing a game agent to play Connect4. I got to explore developing my agent from using a random play strategy to using heuristics to finally using a neural network and the PPO policy to help the agent learn from experience. With this exploration, there was a lot that I didn’t understand about the latter algorithms and I really want to know! However, I know these concepts will be covered further in the deep rl nano-degree so I’m prioritising playing through that. Focus is key to growth!
✅ Sentdex: Intermediate python programming
The was a wonderful overview of more advanced python concepts some of which I have not come across before. The enthusiasm of Harrison who shares his expertise was wonderful via a helpful live-coding style, rate him highly. I especially liked learning about multiprocessing, logging, decorators plus *args and **kwargs. Each concept was only covered briefly so I’m glad to have the familiarity but I’m aware that to gain a decent level of competency, I need to do play through some more code. For this, I’ve discovered that Real Python have some epic tutorials! Looking forward to exploring these in future after I’ve made more progress with the MLSA.
✅ Cloud computing basics
I discovered this mini-mooc on class-central which was highlighting courses that are free during the coronavirus pandemic. Since I didn’t know much about cloud computing but was vaguely aware that top companies like OpenAI using solutions like Azure for training ML algorithms. The brief tour was helpful in giving me an insight into the various options from private to hybrid to public clouds that organisations use depending on what kind of infrastructure they need. Also got to learn about docker importance in creating stable environments and kubernetes for scaling and automating deployment.
✅ Jupyter lab guide
Stumbled across this helpful guide to Jupiter labs which are the next step from Jupiter notebooks. While brief, it’s a great intro to the new functionality which seems much easier to use! I look forward to using it when I resume playing through Grokking deep learning.
✅ How to learn online
As I was reflecting on my progress, I was disappointed in my relative productivity. Some of these failings were structural so I took this brief-course to diagnose areas I was lapsing in. These included issues with an irregular sleeping schedule, not getting enough exercise and building more social time into my routine. This reminder showed me that my health must be maintained otherwise optimal learning will not occur. Note to self: research shows that maximum brain growth occurs only with ample sleep and movement!
What am I still exploring?
Deep reinforcement learning nanodegree
I continue to make progress in the deep reinforcement learning nano-degree. I’ve learned quite a bit about the RL problem from the structure of Markov decision processes to reward structures. I’ve touched upon the theory behind policies, state-value & action-value functions plus how you go about finding optimal policies and am now exploring how monte-carlo methods can help with this. To help consolidate this knowledge, I will be writing an intro post on reinforcement learning that’s focused on the intuitions I’ve picked up! Whilst I have not made as much progress as I wanted, I have managed to gain knowledge which I didn’t have before and am looking forward to making greater strides this month!
What challenges did I face?
My main struggles so far have been with my mind and emotions. I’ve felt many times that I was not able to make progress thought every time I’ve made the time and focused on tenacious exploration, I have broken through to understanding. I need to be more patient with myself and place my awareness on the process instead of the result! I’ve noticed that sometimes, I’ll experience mental apprehension about a new concept because I’ll start thinking about how it’s going to be difficult. I can’t take this seriously! Why? How can I know it’s going to be difficult if I’ve not explored it? Also, everything seems challenging at first before you get started so I again need to give myself the space to play without setting irrational expectations on the journey.
With regards to energy levels and enthusiasm, there have been fluctuations which I’m addressing! How? I’m committing to maintaining healthy habits and watching interesting AI videos to help nurture my internal motivation. With regards to emotional intelligence, I have become more detached by applying the techniques from my meditation practice on a daily basis to keep me balanced! It’s also helping with being more comfortable with not knowing as this will always be present when stepping outside my comfort zone to gain the treasure of knowledge. It’s with this zest for learning that I go forth to explore more of my MLSA curriculum.
What are my next steps?
- Write an introductory post on reinforcement learning to consolidate knowledge
- Start using the #100DaysOfMLCode on twitter to track daily progress
- Update my curriculum page with resources I’ve explored
Pingback: Machine learning scholar adventure: Chapter 2 - Azuremis