Brian Hall, VP, Product Marketing, AWS, sits down with John Furrier & Dave Vellante for AWS re:Invent 2019 at the Sands Expo & Convention Center in Las Vegas, NV. #reInvent #AWS #theCUBE https://siliconangle.com/2019/12/03/aws-revs-machine-learning-train-custom-silicon-new-sagemaker-goodies-reinvent/ AWS revs machine-learning train with custom silicon, new SageMaker goodies Machine-learning models are like pancakes; the first one is usually crummy. In fact, if they fed on models, developers might drop a shirt size before they got one to the table. Speeding up iterations and cutting guesswork as well as cost can expedite an algorithm’s trip to piping hot, accurate insights. ML doesn’t just sit there — it works like crazy inferencing, or making decisions. This can get compute-intensive, and drive up costs, whether they’re live in action or in training. Up to 90% of the cost of ML comes from inferencing, according to Brian Hall (pictured), vice president of product marketing at Amazon Web Services Inc. AWS just announced new instances — called Inf1 — made with custom silicon dedicated to ML inferencing. “We have a brand new set of instances that will reduce costs by up to 90% for people doing inference in the cloud,” Hall said. Hall spoke with John Furrier (@furrier) and Dave Vellante (@dvellante), co-hosts of theCUBE, SiliconANGLE Media’s mobile livestreaming studio, during the AWS re:Invent event in Las Vegas. They discussed Amazon’s custom-silicon instances for ML, its ML studio, and other announcements. (* Disclosure below.) Bringing sight and sanity to tweakfest Things that can fall ML on its sail to the predictive bullseye are many: data sets, different parts to the algorithm, sequences, etc. The difference between success and failure may be just little tweaks here and there. “If that’s all black box, it’s really hard to tell,” Hall said. AWS’ ML platform SageMaker gave them a way to break model-building into simple steps. Now, SageMaker Studio’s Integrated Development Environment and SageMaker Experiments for tracking provide visibility. This allows developers to quickly adjust models and return to testing through inferencing. “This gives me, essentially, a single dashboard for my whole machine-learning workload,” Hall said. “Now I can see how models perform differently based on tweaking variables, which starts making it much easier to explain what’s happening.” Watch the complete video interview below, and be sure to check out more of SiliconANGLE’s and theCUBE’s coverage of the AWS re:Invent event. (* Disclosure: Amazon Web Services Inc. sponsored this segment of theCUBE. Neither AWS nor other sponsors have editorial control over content on theCUBE or SiliconANGLE.)