Leveraging machine learning alternatively by Yeshasvi Tirupachuri


7dfe78ba73229db48137f1337d7cc303_the-thinker-clip-art-images-thinking-man-sculpture-clipart_300-336

Image credit : Clipartfest

Author Pamela Mccorduck writes that the quest for Artificial Intelligence(AI) began with an ancient wish to forge the gods. Of late, the frenzy of rogue AI (robot) systems stealing jobs or worse, posing a threat to humanity piqued. This rise in attention can be mainly attributed to the recent advances in machine learning, a field which grew out of a quest for AI. Machine learning is the art of enabling computers to iteratively learn patterns from data without explicitly programming. Two most important branches of machine learning are supervised learning and unsupervised learning. In supervised learning correlations between input data and output data are used to build a model which can predict the output of an unseen input data. It is commonly applied in scenarios where past data serves as a predictor. A simplest example is housing price prediction based on attributes like the total area, number of bedrooms, number of floors. Unlike supervised learning, unsupervised learning is given only inputs without any corresponding outputs. The goal is to learn the relationships within the input data. A simple application is clustering the input data into some groups with common attributes for example, identifying customer segments. Often unsupervised learning is used first to understand the input data better and then supervised learning is applied further to train a machine learning model. This article presents the current trends and shifts in paradigms brought by machine learning in brief and provides insights into applying machine learning in the areas of agriculture, infrastructure development and academics.

Smartphones are omnipresent and are the quintessential examples of cutting edge applied machine learning techniques. A simple and clear example to explain what machine learning enabled is, how your smartphone handles your photos. They can be categorized by places, faces of people, dominant colors, pets, food etc., and a simple search for a keyword “sunset” or “jack” will bring up results instantly. Other important examples are self-driving cars, personalized medical care using more accurate diagnosis and voice assistant systems like Amazon’s Alexa for the future smart homes.

 

As much fun as it is to bask in the convenience brought by the disruptive technologies I believe that there are other areas that can greatly benefit from advances in machine learning. Consider the case of agriculture in India, which ranks second worldwide in farm output and yet is experiencing a steady decline towards economic contribution. By analyzing data on water, land and equipment availability, marketing and storage needs in relation to production output of a particular region can highlight the main problems which can be addressed precisely, instead of trying to tackle several problems and often failing altogether. Another potential area, I believe, is public investments into infrastructure development. Misappropriation by private parties is a well-known phenomenon and this can be effectively tackled leveraging machine learning techniques. Budget submission, when requesting public funds, is almost always overestimated to facilitate private gain margins. By analyzing the data from the past projects, governments can get better estimates of the precise financial needs for projects beforehand.

 

Machine learning can also facilitate knowledge production. The training activities of PACE (Perception and Action in Complex Environments) Initial Training Network gave me an opportunity to learn about neuroscience. I quickly realized that the amount of data which goes into scientific investigations is vast and is often underutilized. This is related to the fact that funding is often available for limited periods. By amassing data from various scientific projects and by applying machine learning techniques I am convinced that interesting patterns can be revealed. The tricky issues here are data management and privacy. A central organization funded by the European Union to handle all the data from projects in several European member states can provide a means to pursue this idea further to fruition.

 

Overall, although machine learning is not a definitive way to achieve human level intelligence, I believe it can be leveraged to find invaluable solutions for many potential problems prevalent today.

facebooktwittergoogle_pluslinkedinmail

Leave a comment

Your email address will not be published. Required fields are marked *