4 ms·
Like the abstract mentioned, I think this is also a good criteria for helping "level the playing field" for groups with lower budgets so the phenomenon of simpl
by ausbah 7y ago
Like the abstract mentioned, I think this is also a good criteria for helping "level the playing field" for groups with lower budgets so the phenomenon of simply throwing more computational resources at a problem with an existing resource becomes less of a novelty.
- remon 7y agoHm, I struggle to see an upside with levelling the playing field that way. Groups that have the budget to throw huge amounts of resources at problems are still providing important insights. That can happen in parallel to optimising wattage/computational unit. In fact, that can be an almost completely parallel track in AI research.
- currymj 7y agoright now such a parallel track doesn't really exist, at least not on an even footing. as the article shows, there's still a heavy focus on beating accuracy benchmarks, at huge computational cost, in terms of what actually gets accepted at major venues. it can be quite difficult to make the case for a method that has worse performance, but is cheaper. hard to judge without real data but I think such papers are much less likely to be accepted. there are also sometimes demands to replicate very expensive techniques as baselines, which can be onerous for groups with limited resources.
- Nasrudith 7y agoThe upside is probably opportunity cost related essentially like mainframes of yore. Technically they could handle higher throughput than the workstations at the start and even today but once developed they became a dead end compared to more parallel approaches. That sort of quantity vs quality thing has happened a lot throughout history. The "VHS beats Betamax" moments that goes way further back but with more obscure examples.