The discussion has been open for a while now. We know we want Machine Learning (ML) to be able to drive the Artificial Intelligence (AI) and smart automation that we demand in tomorrow’s software -- but how do we build, train and further educate the ML brains we seek to now bring forward?
Computer brains are built with models that are exposed to datasets, which are then trained, tested and enriched over time as they learn what’s right, what’s wrong and ultimately what is altogether culturally and ethically appropriate or not.
Inside Microsoft’s ML brain
As already noted on Forbes, Microsoft used its sometimes physical sometime virtual Microsoft Connect(); event this year to detail the next stage of development in its cloud platform’s Machine Learning services. The Microsoft Azure Machine Learning service is a cloud service that enables developers and data scientists to build, train and deploy machine learning models. The software will reside inside Microsoft’s AI and ML portfolio, which includes the firm’s Cognitive Services brand and the wider Azure ML Studio.
So why this update? Because the trend (indeed, the need) among software application developers seeking to put ML ‘smartness’ into our apps is real, yet the task of doing so is laborsome, very complex and requires a lot of backend grunt work.
What developers are looking for -- and this is not a Microsoft message, this is what the whole industry is talking about -- are routes to be able to abstract that backend complexity and (in the most non-technical terms) plug in chunks of additional ML intelligence into the apps they are trying to build.
According to Microsoft’s Connect(); event ‘book of news’ story compendium (yes, that’s now a thing, really) documentation, “Azure Machine Learning service eliminates the heavy lifting of end-to-end machine learning workflows and can reduce time to production from weeks to hours. It enables data scientists to automate model selection and tuning, increase productivity with DevOps for machine learning -- and easily deploy models to the cloud and the edge."
How was this brain built?
The advancements that now come forward in this part of Microsoft’s ML brain play result in part from experimentation carried out by Nicolo Fusi, who works in the automated machine learning research team at Microsoft Research.
Working on gene editing experiments, Fusi found that trying to work out which ML model to use was too complex and, quite simply, just not a good use of time. His team instead focused on developing an AI capability that could automatically perform the data transformation (categorizing which data is relevant to an intelligence task, how it should be parsed and filed… and in what order of importance it should be ranked), the model selection for the job (what data rules govern what happens inside the AI brain) and the hyperparameter tuning (constraints, weights or learning rates to generalize different data patterns) part of AI development.
His team ended up inadvertently creating the product that has been developed into the new automated Machine Learning functions available to use for Microsoft Azure Cloud platform applications.
Microsoft also announced that the Azure Machine Learning service now includes a software development kit, or SDK, for the Python programming language, which is popular among data scientists.
“We heard users wanted to use any tool they wanted, they wanted to use any framework... and so we re-thought about how we should deliver Azure Machine Learning to those users,” said Eric Boyd, corporate vice president, AI Platform, who led the reimagining of the Azure Machine Learning service. “We have come back with a Python SDK that lights up a number of different features.”
These features include distributed deep learning, which enables developers to build and ‘train models faster’ with massive clusters of Graphical Processing Units (GPUs) and access to powerful Field Programmable Gate Arrays (FPGAs) for high-speed image classification and recognition scenarios on Azure.
Is AI actually getting smarter?
As deeply geeky as the preceding paragraph is, the important part is the ability to train Machine Learning models faster. We need to build AI systems that make the next generation of intelligence happen, but the training part is tough and expensive and prone to errors.
Microsoft’s John Roach notes that, “Automated machine learning homes in on the best so-called machine learning pipelines for a given dataset in a similar way to how on-demand video streaming services recommend movies. New users of a streaming service watch and rate a few movies in exchange for recommendations on what to watch next. The recommendations get better the more the system learns what movies users rate highest.”
It takes a lot of resources do this ML training and expose artificial brains to large sets of data, different application usage and execution scenarios, different platform requirements and rapidly changing functionality demands as users move quickly from one set of requirements to another in the always-on always-connected world of cloud and mobile computing.
If we can train machine brains faster, then we can deploy smarter apps faster on expensive IT equipment that works better first time around is inherently more flexible and interoperable so that it can also be used elsewhere in other applications -- that’s what Microsoft has aimed to commoditize for us here… and it might just be a smart (human, not machine) move.
