Embracing Volatility: Using ML and AWS to Become More Efficient Amid Epic Uncertainty
This blog post summarizes what I learned in a presentation at NVIDIA’s GPU Technology Conference (GTC) by the charming Allie K. Miller, US Head of ML Business Development for Startups and Venture Capital at AWS, titled “Embracing Volatility: Using ML and AWS to Become More Efficient Amid Epic Uncertainty”, as well as my own my own studies and experience.
To define uncertainty, I’d like to use numbers to give you a better idea of what it is, rather than use words. Uncertainty is 2020. I think we all can agree that 2020 has taken the world by surprise, and even toddlers have a fair idea of what it really is. For a toddler though, such uncertainty would be welcomed as he’d be more than happy to see his kindergarten get canceled for 5 months. But for a startup just getting off its grounds or any businessman, uncertainty would give anxiety to many of them.
For a generation of business leaders, volatility has been a part and parcel of them and there's always been normality when it comes to dealing with it. Technology transformation, disruptions, emerging business models, and a developing consensus on globalization have always challenged the very basis of decisionmaking and kept business models on their tip-toes.
The COVID-19 pandemic, however in the past seven months, has pushed every business leader, analytical machine learning model to its extreme, and the threshold to this extreme just seems to increase continually because in our collective calculus of uncertainty, there is no event in history where such a pandemic can be referenced to.
Lets us talk about VUCA. Wait, weren't we talking about volatility? well VUCA was an acronym coined by Warren Bennis and Burt Nanus in their leadership theories during the cold war days, to reflect on how the modern world would adapt after the cold war.
How do we interpret VUCA when it comes to the business world, especially during these never-seen special “COVID-19” circumstances?
Volatility means handling the ups and downs which arise during supply and demand. Not being able to handle this fluctuation leads to loss of confidence and trust in your consumers.
Let us take an example of Instagram. Its a highly know technique ( at least in India ) that people upload their posts around 6:30 pm because most of the users tend to be online in the evenings, usually after 7. Hence Instagram is known to use high compute during these hours. Maximum demand, maximum supply! What if Instagram fails to deliver during such scenarios of high demand? The app crashes and leads to an online outcry, leading to a loss of trust. Well since Instagram is a part of parcel of your life, you wouldn't go uninstalling it. But what if it were a startup or a company trying to increase its outreach amongst its few consumers, well then it would no longer have the people’s confidence and hence it’s own customers might look for other alternatives.
How does Amazon help tackle volatility?
The Amazon provides various platforms through which one can Scale data and business smartly.
- The Amazon Sagemaker helps data scientists and developers to fully build, train, and deploy a machine learning model quickly. It helps in automating DevOps.*
- Managed Spot training to reduce training costs by 90% as the Sagemaker makes the spot interruptions for you.
- S3 intelligent tiering for automatic cost-saving on infrequently used data. For example, what if Instagram sets for high compute at a period of low supply? this leads to a waste of resources and an increase in machine costs. S3 intelligent tiering hence smartly brings balance to the supply-demand chain.
Uncertainty is the unpredictable, surprising factor that comes into businesses. For example, your competitor launching a product just a day before your product launch, and it seems like the public has got favoritism towards your competitor's product, than yours. This leads to delayed agility and makes us get into thinking about how to react to this. Now your business will be slow to react to this change due to unforeseen circumstances.
How does Amazon help tackle uncertainty?
Amazon is always adapting to change, but it's also about the speed to adapt to change. BlueVine was able to process 100000 pages of loans so that they could save about 400000 jobs.
- Amazon Forecast helps us to predict demand. It deals with data from the past 20 years so that it can predict things based on forecasting demand.
- Amazon SagemakerDeepAR algorithm helps us to build our own prediction models.
- Amazon Textract helps to quickly scale workflows with documents.
Complexity in business terms is too much data. Everywhere. Big data. Messy data. Unstructured data. Change in data due to volatile and unpredictable circumstances.
Amidst all this Data, our business model might just end up missing the hidden needle in the data haystack.
Let us talk about CitiBot, which builds webchats, interactive text-alerts, and text chats for Modern Government and Connected Communities. When Covid-19 was at its peak of outreach, the number of calls increased and most of them were about their jobs safety and knowledge about the virus. It was heard that the wait-time got to 3 hours. Well, no one would wait that long for a reply, hence 50% of the people bounced, given the waiting period. The need of the hour required quicker replies to reduce this time. CitiBot managed to reduce this wait-time by 90%.
How does Amazon help tackle complexity?
Amazon significantly cleans the data by distilling information.
- Amazon Lex is used to build conversational AI.
- Amazon Kendra is able to query a large amount of data and process it with Natural language. To give you an estimate, it’ll be able to go through 70000 documents, query our question, and deliver the answer to our question.
- Word2Vec and BlazingText in Sagemaker help us to build our own NLP models. Word2Vec is used to train a large corpus of text through neural networks.
Machine Learning eats complexity for breakfast! ~ Allie Miller
Ambiguity exists when there is more than a single answer to a particular problem statement, and in its own way, they are equally valid. The challenge exists to choose the right path and get an accurate solution in the business world, as this ambiguity could lead to lower conversion rates. What does a lower conversion rate mean? Supposedly you go on Amazon, you search for some shoes you needed to buy. They are tasteless and something you’d never buy. Disappointed, you close the tab. You come across the same shoes in a pop-up ad, and you wonder, why am I seeing such an ad where the product is of no use to me? Clearly, the model is failing to figure out the needs of the person. When a particular solution you propose fails to be the required solution for the consumer, it leads to a lower conversion rate.
How does Amazon help tackle ambiguity?
The main point is to create a market of one.
- Amazon personalize is a trainable API based on 20 years of Amazon experience that helps us to easily create a recommendation engine.
- Amazon Sagemaker and XGBoost help us build our own classification engine.
- A2I also called Augmented Artificial Intelligence, helps us integrate human decision-making with AWS. If the model gives us a particular threshold in a multinomial classifier, for example 40% for a particular type of shoe, then the human can go and control this, based on how much we are iterating.
Well, Allie Miller managed to add her own unique perspective in VUCA, making it VUCAH. The H here stands for humanity. I’d rather just show you what she meant, I don't think I can do justice to her insight and words.
The blank in the end is for you to fill in. I think what she mentioned above could also be used to relate in real-life scenarios. This pandemic has taken us by surprise and I think everyone personally is on a different track as to what they thought they would be doing 6 months ago. During these tough times, life can seem to be getting out of control. But even if you're not in control, you can still drive for it. Hence it is not about scalability on how much has changed, but about the factor of adaptability to these changes in your life.
Two quotes from Allie Miller which are too serene.
Stay wildly optimistic while properly paranoid.
It is just not about planning for scalability. It is also about planning for adabtibility.
Check out the NVIDIA GTC which has brought me to the infinite wandering of knowledge on Deep learning, GPUs, AI, and Machine Learning.
Please do add me on Linkedin and follow me on Medium for more DL, AI blogs!
