First of all, listening to https://soundcloud.com/junojuno2/sets/postinos-house-set right now is fueling my mood, which I am sure will reflect in the tone of this post. Listen to it with me!
A few days ago I attended a Meet-Up that meets monthly at PeopleSpace in Irvine, CA called “Applied Artificial Intelligence and Analytics”. The leaders of the group are working to gather people working and interested in AI (artificial intelligence), and form a community. Below is a description of my experience:
PeopleSpace is a cool maker-space! As you walk in, there are rows of 3D printers, tools, a nerf-gun wall, plenty of open space, and even some houseplants (kept alive with an automated system including a pH monitor, of course). I grabbed a seat and immediately launched into conversation with others attending the meeting. Some attendees were entrepreneurs, some were data scientists, and others were graduating students looking for work. The demographics of the attendees were as I had expected; I was one of two women who were attending out of about 30 people. (Where are all the women on a Thursday night?!)
After a round of introductions, the first presenter discussed his project: Donkey Car. Donkey Cars are autonomous RC cars! Visit the link for more details.
There are communities who build these cars and race them! It seems like such a neat project to learn AI, and, specifically reinforcement learning (RL). RL means the car teaches itself, so you don’t have to train the car for up to 30 minutes!
The presenter also showed a similar league to the Donkey Cars: AWS Deepracer League. Hosted by Amazon, this league has tournaments too that seem to be like Nascar for programmers.
(Also, where are all the women in this league? Maybe I need to sign up.)
These autonomous RC cars are just so awesome! And, on that note, Tesla does have little cars. This may be the only Tesla I can afford, but I am not sure its range cannot get me to work and back. Maybe I can make that mini-Tesla autonomous as a side project! (haha)
Back to the event:
The second speaker gave the Graves-Ransato presentation from NeurIPS 2018. The presentation was an introduction to self-supervised learning (which means that the computer teaches itself). While this technique is incredibly powerful and saves a lot of time, it is difficult to assess why a computer outputs a certain answer. There’s a push for self-supervised learning, but we need to know why AI made the decisions it did. For instance, if your AI outputs that you need to make X business decision, how would you know the associated level of risk? Why X and not Y? Currently, there’s a big need for a language so that AI can tell us why it decided X, and not Y or Z.
The presentation also discussed Neural Network models. An example of a neural network model is the auto-text we have in our messaging app on our phones. Based on previous words, and the data the computer has on what people typically type in a sentence, the computer suggests what your next word may be!
Recurrent Neural Network (RNN) models have loops, where multiple copies of the same network are in series, each passing a message to its successor. This is good for modeling sequences and lists. To learn more the speaker suggested reading “The Unreasonable Effectiveness of Recurrent Neural Networks”.
What really caught my attention was WAVENET!!! WaveNet is a deep neural network for generating raw audio waveforms (so, it’s sound and tech, of course I’m interested!). This technology covers anything from speech to music!
PixelRNN is another technology that generates pixels using deep neural networks. It’s crazy how close the computer comes to guessing the original image (completions) from the occluded image!
After the presentation, my main question was: how to get started learning more!
Places to start:
- Choose a neural net:
- After choosing one, start with artificial neural network or start with a decision forest.
After the talks, I got to speak more with the presenters and attendees! Everyone was so welcoming and friendly! It was a great intro to AI, and I look forward to attending more events like these in the future.