AI with AI
Episode 3.16: Where NOMARS Has Gone Before
Andy and Dave discuss the President’s 2021 Budget Request, which increases funding for AI but decreases funding to science in general. Google releases Jigsaw, a tool to spot faked and doctored images, but not for the public. DARPA announces its NOMARS program, a No Manning Required Ship. Google creates an ML “fairness gym” to let researchers explore the long-term paths of AI’s decisions. The U.S. Army introduces Aided-Threat Recognition from Mobile Cooperative and Autonomous Sensors (ATR-MCAS), to assist soldiers in using suites of sensors on the battlefield. The Army also issues an RFI for a Sense-Through-the-Wall System. In research, Facebook AI demonstrates the ability to use ‘radioactive data’ to detect if a data set was used for training a particular classifier. PlosOne and the University of Liege in Belgium graft a neuromodulation capability onto deep neural networks as a way to learn adaptive behaviors. Marek Rei has collected a database of ML and NLP publication statistics with an interactive interface. MIT’s Lincoln Lab disseminates AI: A Short History, Present Developments, and Future Outlook (originally published but only internally disseminated a year ago). An Introduction to Machine Learning Interpretability by Hall and colleagues is the book of the week. Mary “Missy” Cummings pens a thought piece on rethinking the maturity of AI in safety-critical settings. The 34th AAAI publishes a video of the winners of the ACM 2018 Turing Award: LeCun, Hinton, and Bengio. And Denis Shiryaev uses a variety of techniques to upscale and colorize an 1896 short film to 4k and 60 frames per second.