Home >
Introduction to Machine Learning and Deep Learning
Peter McLaughlin - Watch Now - Duration: 36:33
Hello, Chess being relatively simple compared to Go, it can be solved with a rules-based approach. A Deep Learning based system would most likely out perform a rules-based system but it's much more work to build the dataset. In our computer vision work, we generally stick to rules based algorithms if we can for simplicity.
Second /comment/question: before machine learning became popular, I remember the term "fuzzy logic" was popular. Are you familiar with that, and can you tell me what that was about? I remember thinking, at the end of the day, it was just a more complicated decision tree in your program than simply "is it equal to 1"? At that point, it did not seem like anything noteworthy to me. Was I mistaken?
Hello Fuzzy Logic still has to be explicitly defined rather than being trained from a data set. It is therefore easier to modify than Machine Learning / Deep Learning but it can't tackle as complex problems. It is still heavily used in control system decision making.
Hi, great talk! What kind of software or libraries are you using while dealing with machine learning especially deep learning? Cheers!
Hello, the main libraries are TensorFlow (and Keras which wraps TensorFlow), PyTorch and Caffe. Python is typically used for training and C++ for inference.
And my third (and final?) question: In the "Machine Learning vs. Deep Learning" slide at 24:53, you have the neural network type diagram under the deep learning. To give me something concrete, could you give some simple actual examples of what might be in a few of those decision bubbles?
Thanks again, it was the best introductory exposure to machine learning vs deep learning I have had so far.
Hello, the nodes of a neural network contain weights for each of the inputs to that node, followed by an "activation function" which determines the node's output value. Here are some examples of activation functions: https://en.wikipedia.org/wiki/Activation_function.
Thank you for watching. Apologies if the audio quality was bad at times.
Will the presentation charts become available soon?
Had issues hearing the speaker. Someone was speaking in the background.
Do you mean in the video itself or in your own physical environment?
In the video itself. To verify, I paused and unpaused several times and the background voice tracked. It didn't seem like part of the presentation as it was interrupting. I could be wrong. I was trying to listen in the background. Maybe the speaker was using audio for his input?
Let me know if you ever locate the timestamps for these, I'm curious. Thanks.
I will re-listen later this afternoon and let you know. Did you have a chance to read my email about access to the 2020 conference?
I relisten to the section I heard the background talking. It was about 32min into the video. However, relistening to the video at that point the background talking was not present.. It should similar to what you would experience if someone had not muted their mic. in a Zoom presentation. However, since this was not Zoom, and pre-recorded, I have to assume 1) an abnormality, or 2) I heard a voice in my head? Hard to believe, but pressing the pause/play and having the voice track, maybe some metaphysical connection? :~V
And can you provide the slides for download?
The Q&A for this session had to be canceled. Please feel free to ask questions here.
Thanks, excellent talk, and very well suited to my introductory level. First question:
Since we have much more computing power and familiarity with machine learning today, if we had the same situation as in 1996, where people were trying to write software to beat Garry Kasparov, would they decide that rules-based programming is too much work, and opt for machine learning instead?