Gábor Melis is a consultant at Franz Inc. He worked on the SMP thread implementation of both SBCL and AllegroCL and got interested in all things parallel. Had he been a first generation lisper, he would have been in the artificial intelligence bubble of the 80s, but as a second generation lisper all that was left for him is machine learning. There are two kinds of programmers: those who get a kick out of their creation being used by real people to accomplish a goal, and those who first and foremost think in terms of the internal beauty of the object at hand. Most of the time, Gábor is firmly in the latter category and for his practical side to emerge he needs external pressure. Fortunately, external pressure is abundant in contests which he finds and leverages with some success.
Talk — Sending Beams into the Parallel CubeA pop-scientific look through the Lisp lens at machine learning, parallelism, software, and prize fighting
We send probes into the topic hypercube bounded by machine learning, parallelism, software and contests, demonstrate existing and sketch future Lisp infrastructure, pin the future and foreign arrays down.
We take a seemingly random walk along the different paths, watch the scenery of pairwise interactions unfold and piece a puzzle together. In the purely speculative thread, we compare models of parallel computation, keeping an eye on their applicability and lisp support. In the the Python and R envy thread, we detail why lisp could be a better vehicle for scientific programming and how high performance computing is eroding lisp's largely unrealized competitive advantages. Switching to constructive mode, a basic data structure is proposed as a first step.
In the machine learning thread, lisp's unparalleled interactive capabilities meet contests, neural networks cross threads and all get in the way of the presentation.Back to the symposium programme.