Am I training my Machine Learning models or are they training me?

Source: Deep Learning on Medium

Source: http://scyfer.nl/2017/06/19/active-learning-human-in-the-loop-ai/

Am I training my Machine Learning models or are they training me?

3 Critical Life Lessons I learnt from training Machine Learning models

As a Data Scientist, spending hours and days working on Artificial Intelligence, I could not help but philosophize about Machine Learning. As I delved deeper into the core ideas that form the basis of this marvelous technology, I uncovered the following three life lessons that I believe to be fundamental to human progress and success.

1. Reaching life goals by ‘backpropagating’ through mistakes

The backpropagation algorithm is undoubtedly one of the most fundamental building blocks of Deep Learning. The deep-seated intuition behind the algorithm is a very fundamental human idea — learning from mistakes.

“By seeking and blundering we learn.” — Goethe

During the training phase of building a Deep Learning model, when the model makes a wrong prediction, the backpropagation algorithm adjusts the parameters of the model marginally, so as to steer the model towards making a ‘more correct’ prediction. One can think of these parameters as the dials on an FM Radio which are adjusted back and forth till the desired radio station can be heard clearly.

Source: https://s3-eu-west-1.amazonaws.com/

When pursuing our goals, we often adopt a self-defeating negative attitude towards our errors, mistakes and failures. Mistakes often seem to make the goal appear farther and even unattainable, when in reality one might just need to treat mistakes as a feedback signal and adjust one’s approach accordingly. The famous life-coach Tony Robbins popularized a similar idea — the 2mm rule, which is akin to the notion of learning from mistakes and making marginal adjustments to one’s approach in order to progress towards one’s goals. Perhaps, we can learn to perceive mistakes with the same dispassionate and goal-oriented perspective with which a Deep Learning model betters its performance. By backpropagating through our errors, we can make our mistakes the very basis of our imminent success.

2. Avengers…..ENSEMBLE! Teamwork makes the dream work

Source: Avengers Endgame

Ensembling is a fundamental method for enhancing the performance of a Machine Learning model. It is a technique where multiple models are combined to improve overall performance, which can generally not be achieved with a single model alone. Much like having a team solve a particular problem together, ensembling generally yields improved predictive power.

“Alone we can do so little; together we can do so much.” — Helen Keller

While ensembling comes in various flavours such as bagging, boosting and stacking, the core idea behind these techniques is to use a combination of models to buttress each other’s strengths, while attenuating the insufficiencies of individual models. Similarly, putting multiple minds together to solve problems has been a tried and tested formula for success. In the world of business and corporations, teamwork is given so much importance that companies like Google are spending large sums of money to figure out how to assemble the perfect team. The paramount importance of teamwork is conventional wisdom, if not a cliche notion. However, conventional wisdom and cliche notions have a propensity to be ignored, forgotten or taken for granted. Especially among techies like myself, there is a tendency to want to avoid communication and work on a particular problem in silos. Asking a fellow developer for help can be considered as a sign of incompetence and God forbid if a senior engineer asks a junior engineer for advice! However, we must remember that just like ensembling, where multiple Machine Learning models cover each other’s flaws, working in teams is a pathway to both personal and team success.

3. Garbage in, garbage out

Source: https://i.pinimg.com/originals/b6/5f/04/b65f044e766a0f2fe8ad531eeee8e6a0.gif

A Machine Learning model is only as good as the data it has been trained on. If the data is corrupted, the resulting Machine Learning model will be corrupted. If the data is biased, the model will produce biased results. The importance of the quality of the data being used to train a Machine Learning cannot be emphasized and iterated enough.

“Every day, stand guard at the door of your mind” — Jim Rohn

Today, massive volumes of information — good, bad, false, useful, useless, harmless, harmful— make voracious attempts to seize our mental bandwidth every single second. Now, more than ever, it is crucial to critically examine the velocity, volume, value, variety and veracity of the data to which we expose our minds. This topic has been profusely discussed and written about in the news, articles, books, blogs and journals. In his book ‘Digital Minimalism’, best-selling author Cal Newport warns us about the far-reaching ramifications of information overload and the addictive behaviour of continuously engaging with ‘rubbish’ online content. Just as a Machine Learning model is limited by the quality of the data it is exposed to, the human mind too is limited by what information it is exposed to. For the sake of superlative decision making, tranquility of the mind and an overall rich life experience, it is indubitably crucial to be wary of the quality of information to which one is exposed.