“Most important thing is to sensitise customers who are new to ML with the ML results that vary from fixed results that we observe in software development.”

For this week’s ML practitioners series, we got in touch with Amit Deshpande, Senior VP at SpringML. Amit’s interest in algorithms was triggered during his time at Texas Tech University where he was pursuing Masters. He oversees the operations of SpringML’s Indian branch. SpringML is a Google Cloud Platform premier partner with specialisation in Machine Learning and Data Analytics. In this interview, Amit shares snippets from his journey in the world of data science.

AIM: How did your data science journey begin?  Your fascination with algorithms. How did it all start?

Amit : It all started with the famous Andrew Ng course on Machine Learning a few years ago. I was following the trends in Machine Learning but this course really explains the low level detail of what exactly is happening under the hood in Machine Learning. You really appreciate the Machine Learning Algorithms or their usage and what’s happening better after going through those nitty gritties which he teaches us. 

AIM: What books and other resources have you used in your journey?

Amit : I used to read up on various articles on medium or towardsdatascience. I believe what has also helped immensely is all those articles on Kaggle. They specifically tell what variety of projects exist and also various insights into the approaches that can be taken. 

SpringML, being a Google Cloud partner, heavily uses TensorFlow and the AI/ML products provided there. The resources available from GCP and the products have helped a lot.

AIM: Tell us about your role at your current company. What does a typical day look like?

Amit : I am the Head of the SpringML India Development Center. My role is to ensure our customers are successful and our employees are happy. My typical day starts with a cup of lemon tea and honey over which I tend to check my emails and calendar for the day. I then take a walk for about 45 mins to one hour, it’s a good time for introspection and sorting the activities for the day. On alternate days, I also tend to meditate for 15-20 minutes. I then proceed with overseeing the customer projects in progress, ensuring they are progressing smoothly and making adjustments as necessary. I am involved in the technical discussions – architecture, design or sometimes implementation. I also keeping a watch on the hiring activity. I speak with almost every potential new joiner to our company. We are hiring aggressively and I try to make sure that our hiring goals align with our targets.

AIM: How does your team approach any data science problem and what challenges did you face?

Amit : We have had many Machine learning projects in the last three years. Almost always the approach involves a strong focus on understanding the business problem – the domain knowledge. Getting a thorough understanding of the domain, the factors involved in influencing the model must be the first step after which one should ensure thorough Exploratory Data Analysis. The next most important thing if the customer is exploring Machine Learning problems fresh is to sensitise customers that machine learning results vary from what we observe in software development where the results are pretty much fixed. The end result depends on the accuracy of the model which in turn depends on the data. Once this is established it’s important for our team to plan thoroughly and keep buffers in place to handle the potential delays resulting from re-running the model training due to new factors introduced or changes in data. The standard procedures otherwise hold i.e. creating training, test data, choosing the correct ML algorithm, depending on the problem at hand. 

The initial problems were around compute power on laptops but that was taken care of by the Google cloud platform. Over the years we used ML to solve many real world problems. For instance, in Memphis, approximately 32,000-man hours are spent every year to repair potholes. Memphis Mayor Jim Strickland and CIO Mike Rodriguez began looking for ways they could apply technology to fix the problems. Mike approached Google for ideas, and Google recommended conducting a machine learning proof-of-concept (POC) with SpringML. The POC began by training TensorFlow models for ML object detection using preconfigured AI Platform Deep Learning VM Images on Compute Engine. SpringML helped set up cameras and developed a user interface to collect pothole data and automate the 311 ticketing process.

Together, the teams analysed 30 days of video from a moving city bus and high-resolution video from 360-degree cameras mounted to a code enforcement vehicle, overlaid with data from 311 reports. As the models were refined, accuracy quickly climbed from 50 percent to over 90 percent as models were taught to differentiate a pothole from a manhole cover or other object.

SpringML was able to cut down over 20,000 man hours in predictive maintenance of potholes for the city of Memphis. Another use case was to accurately predict vehicle downtimes by analysing telematics data from over 1.4 million vehicles for a prominent company. More details of the SpringML machine learning use cases can be found here.

To be frank, initially it felt like almost every problem was impossible to solve. But we did solve almost all of them. It requires a lot of perseverance as you will not be able to google the answers. 

AIM: What does your machine learning toolkit look like?

Amit : Being a partner of GCP and TensorFlow, we use the frameworks/libraries available via these two platforms. Python is our choice of language for all machine learning related programming and the popular libraries like scikit learn, pandas, numpy etc are used extensively. Those interested can go through my TensorFlow blog here.

AIM: There is a lot of hype around AI/ML. Which domain of AI, do you think, will come out on top in the next decade?

Amit : We have been in this space for a while now and I strongly believe at this stage we can no longer say there is a hype around AI/ML. This is real and we must believe in it and make the best use of it. Combined with hardware, AI is going to revolutionise the way we think and work. AI/ML is solving real world problems which were unimaginable a few years back. Take the case of a media company having a few dozen photographers who take tens of thousands of images as part of their job. Sorting through these images in the digital era is going to be extremely time consuming. With AI/ML, we were able to categorise images based on city, people, was it a sunny day or was it snowing and 100+ such categories. This has resulted in amazing efficiency. The same way we are able to categorise whether fruits are ripe or not, we can now club this trained model to pick the fruits using a drone. At this point, AI/ML has taken off but the real take off will happen when the technology is well digested by many and creativity kicks off. AI is enroute to going mainstream soon.

AIM: What would you advise aspirants who want to crack data science/ML roles at your company?

Amit : First and foremost,it sounds exciting and cool, but be ready for a lot of grunt work and patience. Secondly, a clear understanding that ultimately to get into these roles you must have solid programming skills as well. When it comes to learning, Coursera and Kaggle are great resources.