The definitive weekly newsletter on A.I. and Deep Learning, published by Waikit Lau and Arthur Chan. Our background spans MIT, CMU, Bessemer Venture Partners, Nuance, BBN, etc. Every week, we curate and analyze the most relevant and impactful developments in A.I.
This week we cover Allen Institute's Project Alexandria which got Paul Allen's $125M injection, the goal of the project is to teach machines common sense. We discuss what it entails and whether it can be a successful effort.
In our other sections, we link you to Google Machine Learning Crash Course, which is a set of 15 hours to teach you basics of deep learning.
As always, if you like our newsletter, feel free to forward it to your friends/colleagues!
This newsletter is a labor of love from us. All publishing costs and operating expenses are paid out of our pockets. If you like what we do, you can help defray our costs by sending a donation via link. For crypto enthusiasts, you can donate by sending Eth to this address: 0xEB44F762c58Da2200957b5cc2C04473F609eAA65.
You may heard this week that Paul Allen will inject another $125M dollars into his own Allen Institute. The goal is to teach common sense into AI. So what's the deal? How's this effort different from our current deep learning approach?
A video from Allen Institute, narrated by CEO of Allen Institute, Oren Etzioni, explain it well. What Etzioni and his colleagues try to do is use crowdsourcing and ask a large number of people on various common sense questions. Once the answers of these questions are collected, the team will use them to inform various specific AI projects. They call this database of common sense questions as Project Alexandria.
Will this approach work? We think it is a huge step forward from projects such as Cyc which experts was hired to curate relationship between different concepts. (The fancier term of these relationships is ontology.) Alexandria focuses on common sense which we believe is more important in real-life applications.
Would these rules be useful in specific AI ? That depends on the type of tasks. Notice that the answers of these common sense questions are likely to be either natural language response from humans or categorical responses (such as an object name). So the success of the project depends on whether researchers can leverage natural language or categorical response in adapting the existing AI algorithms.
All-in-all, this is a worthwhile effort which expands our current AI technology. Let's wait for the good news from the Institute then?
As the Google TPUv2 Cloud is released, we start to see benchmarking results. Here is a post from RiseML's Elmar Haußmann, who benchmarked the performance of TPUv2 vs GPU. He found that TPU performs well favorably compared to GPU both in performance and economically.
This is a dataset shared by Google recently on landmark recognition. Unlike the standard Imagenet task, landmark recognition has the inherent class imbalance issue, i.e. some of the lesser-known landmarks might have fewer training data. So that is perhaps why landmark recognition could be an interesting domain, and it becomes a task at CVPR2018.
Google just released a crash course with 15 hours of material. We took a quick look, the course is more introductory material which is suitable for beginners. More interestingly though, the videos are more tailored to engineers who like to deploy a TF models, which we found a very helpful in practical deep learning development.
The Weekly usually cites links of the latest research/resources. But occasionally we feature our own AIDL members' work. In this case, Dibakar Saha creates an impressive demos of sign recognition and calculator, which attracts a lot of eyeballs.
This newsletter is published by Waikit Lau and Arthur Chan. We also run Facebook's most active A.I. group with 110,000+ 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 with this iOS app here: https://itunes.apple.com/us/app/expertify/id969850760
Artificial Intelligence and Deep Learning Weekly
Speech Recognition, Machine Learning, and Random Musing of Arthur Chan