We consider interactive learning in the realizable setting and develop a general framework to handle problems ranging from best arm identification to active classification. We begin our investigation with the observation that agnostic algorithms \emph{cannot} be minimax-optimal in the realizable setting. Hence, we design novel computationally efficient algorithms for the realizable setting that match the minimax lower bound up to logarithmic factors and are general-purpose, accommodating a wide variety of function classes including kernel methods, H{\o}lder smooth functions, and convex functions. The sample complexities of our algorithms can be quantified in terms of well-known quantities like the extended teaching dimension and haystack dimension. However, unlike algorithms based directly on those combinatorial quantities, our algorithms are computationally efficient.

We consider interactive learning in the realizable setting and develop a general framework to handle problems ranging from best arm identification to active classification. We begin our investigation with the observation that agnostic algorithms \emph{cannot} be minimax-optimal in the realizable setting. Hence, we design novel computationally efficient algorithms for the realizable setting that match the minimax lower bound up to logarithmic factors and are general-purpose, accommodating a wide variety of function classes including kernel methods, H{\"o}lder smooth functions, and convex functions. The sample complexities of our algorithms can be quantified in terms of well-known quantities like the extended teaching dimension and haystack dimension. However, unlike algorithms based directly on those combinatorial quantities, our algorithms are computationally efficient.

It's been four full days since Sam Altman's shocking dismissal from OpenAI, and we still have no idea where he's going to land. There are suggestions that Altman, one of the most powerful figures in AI, could return to the company if the board changes significantly--talks are reportedly under way. But there is also an offer on the table from Microsoft to start a new AI research group there, which would be a cruelly ironic outcome for OpenAI, which was founded as a nonprofit with the goal of drawing talent away from Silicon Valley's biggest companies and developing AI safely. How Altman got to this moment is telling. In the days after his firing, he managed to prove that he is far more than a figurehead, winning over a majority of OpenAI employees (including Ilya Sutskever, the company's chief scientist and the reported architect of his dismissal--it's, uh, complicated) and some of the tech industry's biggest luminaries.

In the recent podcast, "Can We Upload Ourselves to a Computer and Live Forever?", Walter Bradley Center director Robert J. Marks and computer scientist Selmer Bringsjord discuss whether we could achieve immortality by uploading our minds to computers. The year 2029 is the consistent date I've predicted, when an artificial intelligence will pass a valid Turing test -- achieving human levels of intelligence. "I have also set the date 2045 for singularity -- which is when humans will multiply our effective intelligence a billion fold, by merging with the intelligence we have created. Indeed, when Kurzweil (left) became a director of engineering at Google in 2012, he not only mainstreamed the basic idea but he "heralded, for many, a symbolic merger between transhumanist philosophy and the clout of major technological enterprise." (The Guardian, 2017). Beyond the Valley, the project gets more ambitious. In a recent piece at Gizmodo, Toronto-based writer George Dvorsky advocates uploading our minds to supercomputers somewhere in the universe, a proposal he calls Distributed Humanity: "Entire civilizations could live on a single supercomputer, enabling the existence of potentially trillions upon trillions of individuals, each of them a single brain emulation.

Every single person who contributed is a true believer of creating a network of compute nodes that utilize idle compute power for the purposes of AI training where speed is the key. These are people who roll up their sleeves in the industry and see a clear path to user adoption when most believe Amazon, Google, or existing frameworks get the job done. Contrarian thinking is how massive opportunities are captured. We are honored to have on-board futurists who are aligned with our vision. While existing deep learning distribution methods and frameworks have come a long way, it still suffers from slow training speeds and expensive servers.

When Cisco started working on a unified analytics platform, the company was more concerned with finding the right people to do the work than the data they were using, Cisco's vice president of strategy Rajat Mishra said at the VB Summit in Berkeley, California today. "I wish I could tell you we had a very analytical approach to this problem," he said in an onstage interview. "The truth is much messier." The management approach Cisco took nicely mirrors one method of training a machine learning algorithm: Start with a small data set, then add more data. "Instead of focusing on data, we actually focused on the people," Mishra said.