If you haven’t caught it yet, be ready to! The hype around ML is so palpable. ML algorithms have been around since the 1950s. So why is ML so hot now? The internet age and the prominence of cloud based platforms allow for collection, storage and analysis of huge amounts of data (the global datasphere is expected to grow 10x to 163ZB by 2025). Additionally, we now have virtually unlimited storage and compute power (we hear of Google’s TPUs and Nvidia’s new Volta with its next-gen GPUs) to accelerate new ML models. You also don’t have to buy a datacenter: you can rent supercomputing power by the minute. Furthermore, there have been breakthroughs and trends in the past 10 years around creation of new algorithms, deep learning applications, and hyperparameter optimization. The tools have gotten better and easier to use across these areas. Finally, free resources are available and accessible from anywhere in the world, and there are growing ML communities like Kaggle where statisticians and data miners can compete to produce the best models for predicting and describing the datasets uploaded by companies and users.
Many of you, just like me, were probably fascinated by the story of the Japanese cucumber farmer who used an inexpensive Raspberry Pi and Google’s open source deep learning library TensorFlow to automate the sorting of the cucumbers. For me, the fascination doesn’t stop there. Today, I get to work with a talented team of engineers, and business strategists who relentlessly challenge the boundaries of what is possible in the field of ML for our customers. Our use cases span the fields of healthcare (disease detection, preventative treatments), retail (customer retention), energy (drilling efficiency prediction), agriculture (irrigation optimization) and manufacturing (quality analysis). One story that carries particular impact is that of a pediatric hospital where the outcome is to reduce instances of severe hospital-acquired infections. We were tasked with identifying the root cause of an infection outbreak that had spread to a number of critically ill patients. Within 4 weeks, we were able to identify the root cause of the infection. With the use of ML, we are able to predict a patient’s day-to-day risk of acquiring the infection and identify the factors contributing to this risk. The model currently predicts 75% of true positive infection causes up to 3 days in advance of normal infection detection allowing clinicians time to proactively respond and potentially prevent these serious infections.
So what does it take to utilize Machine Learning for your enterprise?
First, you will need a source of data and you will need to manipulate them in order to understand them. These raw data could come from mobile, social media, or sensors in IoT devices. This massive amount of information is processed by a Data Ops team, consisting of data engineers and data product managers who transform the raw data, into modeled data, in a large data lake. The Data Science team with knowledge of algorithmic techniques will experiment and train different models, ensembles of models, or neural networks. They have workbenches with different languages (Python, R) and libraries (MLlib, scikit-learn, MxNet, TensorFlow) to generate the models. Once a model is selected, don’t stop there! Data Scientists then turn over the models to DevOps, who can automate deployment and instill a governance process to ensure that the model’s performance remains within the expected range of accuracy and performance, and that the model is retrained periodically. The ultimate goal is to automate business processes, enrich and enhance existing products, and invent and develop new business models or services.
As you can see, it takes the combination of data engineering, data science and DevOps expertise to fully execute on the technical aspect of ML. But what is even more important is the guidance of business strategists with specific industry knowledge to ideate on ML capabilities, using a user-centered design perspective. While the top 3 cloud providers (AWS, Azure, Google) have entered the AI race with each of their own technical differentiations, we, at Pariveda, provide not only the technical know-how, but also the essential vertical expertise to speak the language of your business. Whether you are the head of research, or design, or technology, or finance, or legal, we can help you get the most out of machine learning.
Article originally posted on Sophie Zugnoni’s LinkedIn page