Thoughts From Your Humble Curators
This week we tell you the story of how Google's Venture is using ML to invest, and how AI is used in World Cup 2018.
This newsletter is published by Waikit Lau and Arthur Chan. We also run Facebook's most active A.I. group with 159,000+ members and host an occasional "office hour" on YouTube. To help defray our publishing costs, you may donate via link. Or you can donate by sending Eth to this address: 0xEB44F762c58Da2200957b5cc2C04473F609eAA65. Join our community for real-time discussions here - Expertify
How to build a ML portfolio if I don't have a Supercomputer?
Q: (Excerpt) I am a college student. I have gone through a lot of machine learning materials (.....) But at times it feels pointless because I feel like I never really will have the ability to train these models. Can anyone give me a guide for building a machine learning portfolio?
A: Very good question. Just on the part about experience in training : for most MLE positions, most employers would examine your analytical capability, rather than you experience in using machines. Latter matters but it is your thinking and problem-solving skill which could attract good jobs the most.
If you do want to accumulate good experience in ML training while you are jobless, here are couple of means I found useful:
1, focus on algorithms which allow you to train model with CPU or low-end GPU. e.g. libsvm, liblinear are rather fast package which you can use to train good models.
2, work on smaller problems. Indeed, large problem often entails more difficult issues. But often 95% of ML issues also occur in small problems. So get yourself to be very familiar with them, and always try to generalize them in your thought. Also try to optimize your program in small problem. Those experience often transfer to bigger problem.
3, work with groups which has slightly more computation than you. So it could be your local interest group, it could be an academic group in university, it could be you just save a little money to buy a cheap card such as GTX 1060 etc.
But don't give up. One thing to share: as sorta an old-timer, one of us (Arthur) started his first HMM experiment with Pentium 500, and learn deep learning few years ago with a 10 years old Dell machine. (It took 3 days to train convnet model for Mnist though, lmao. ) So things are possible, you just have to come up with solutions creatively.
This newsletter is published by Waikit Lau and Arthur Chan. We also run Facebook's most active A.I. group with 159,000+ members and host an occasional "office hour" on YouTube. To help defray our publishing costs, you may donate via link. Or you can donate by sending Eth to this address: 0xEB44F762c58Da2200957b5cc2C04473F609eAA65. Join our community for real-time discussions here: Expertify