Learning Problem Solving from ML Algorithms

Hey There,

Did you know that ChatGPT is powered by AI algorithms? While AI gets all the attention, algorithms play a crucial role in making it work. Think of it like a delicious cake - AI is the cake itself, while algorithms are the recipe that determines the ingredients, steps, and order to follow for the best results.

After doing a machine learning 101 crash course, what I learned was fascinating!  I was surprised to learn that the methodologies and mindset are more transferrable than I thought. Each algorithm is actually an approach to solve a problem and obtain an optimal solution, considering different priorities and constraints, such as computation power and time. But here's the best part: these methodologies and mindsets can be applied in our daily lives too!

And given that I'm definitely not a math person, I guarantee that I won't bore you with formulas. Instead, my goal is to extract the generalized methodology and make it applicable to everyone.

To ensure the information I present is accurate, I sought help from my friend and colleague at Accenture, Ben Lo, who has a solid background in mathematics and statistics. So, a huge shoutout to Ben for his invaluable help! And let’s dive in right now!

1. Finding the best solution that is both doable and close to reality.

When designing algorithms, it's critical to balance reality and solvability. A neural network with many hidden layers may be close to reality but too complex to implement, while a linear regression model that assumes a simple relationship may be easy to interpret but not always accurate. Just like baking a cake, you want to follow the recipe to make it taste good, but you also need to be able to actually make the cake. So when solving problems, the optimal case will be finding a solution that's both doable and close to reality.

Imagine you are a budget manager for all the sports clubs in your company for the upcoming fiscal year, and you have been given a budget that is not enough to fulfill everyone's needs. What would be the optimal solution?

The most intuitive approach might be to divide the budget based on membership size, but this approach is too simplistic as it doesn't consider the nuances of costs for different types of sports. For example, sports like horse riding or ice hockey are more expensive than sports like badminton or running.

We might as well consider all relevant factors and implement a long list of procedures, such as checking for the cheapest courts and rental prices of gear for each type of sport and considering the size of each association's membership. However, such an approach would be overly complicated and time-consuming for the administrative staff to implement, and bear in mind that administrative costs could be a significant source of expenses as well.

Therefore, a better approach might be using a weighted allocation method — which assigns a weight or factor to each sport based on its level of need, using a 1-5 point scale for equipment and facility costs. By considering these factors, the model can approximate the actual needs of each association while remaining manageable for administrative staff. The weights assigned can also be adjusted based on changing circumstances or priorities, making the model adaptable to different situations.

2. Divide and Conquer complex problems into smaller manageable problems

Have you ever wondered how data engineers tackle complex problems? One common strategy they use is called "divide and conquer". This approach involves breaking down a large problem into smaller sub-problems that can be solved independently. Each sub-problem is then solved recursively, and the solutions are combined to solve the original problem. This approach can help reduce the complexity of a problem and make it easier to solve. Essentially, it's like breaking a big problem down into smaller, more manageable parts and solving each part separately.

When faced with a seemingly complex business problem in management consulting, the "divide and conquer" approach can be applied just like in data engineering. For example, imagine you are advising a local craft beer brand in Hong Kong on whether or not to expand to the millennial market. Before conducting extensive research, it's important to identify the specific questions that need to be answered to make an informed decision. The problem can then be broken down into four main parts: understanding the customer, analyzing competitors, evaluating the product, and assessing the company.

• Customer: Who are the target customers in the millennial market? What are their preferences, needs, and behaviors? How can the brand appeal to them?
• Competitors: Who are the main market players in the craft beer market in Hong Kong, especially when it comes to targeting millennials? What are their strengths and weaknesses? How can the brand differentiate itself from its competitors?
• Product: What are the current offerings of the brand? What changes or new products should be considered to appeal to this demographic?
• Company: What are the brand's strengths and weaknesses? What resources and capabilities does it have to support the expansion into the millennial market? What are the potential risks and challenges of this expansion?

Breaking a complex problem into smaller, more manageable parts might sound simple, but it's easily overlooked. Just like a puzzle, solving a big problem piece by piece can make it less daunting and increase your chances of success. By breaking down the problem into specific areas, you can identify challenges and opportunities in each one and develop a strategy that works for your situation. It's like being a detective, piecing together clues to solve a mystery.

3. Simulate various situations using Monte Carlo Simulation

Imagine you're trying to predict the winner of the World Cup. Sure, you might have an idea of which teams are more competitive based on past performance and superstar players, but how can you make a truly scientific prediction? Enter Monte Carlo Simulation – a powerful tool for dealing with uncertainties in any field, from finance to sports.

What is Monte Carlo Simulation, you ask? Well, it involves running simulations with randomly generated input parameters within a given range and then averaging the results to estimate the expected outcome. It's like trying out all the possible scenarios and seeing which one is the most likely to happen.

Let's say you want to predict the winner of the 2022 Qatar World Cup. Here's how it works:

1. Identify the variables that could affect the outcome, such as team strength, player injuries, and weather conditions.
2. Assign probability distributions to each of these variables, based on historical data or expert opinion. For example, you might assume that a team has a 60% chance of winning if their star player is healthy, but only a 40% chance if he's injured.
3. Run a lot of simulations, each with a different combination of variables. For example, you might simulate a match with a strong team and good weather, but with one key player injured.
4. Record the outcome of each simulation (i.e., which team won), and then calculate the overall probability of each team winning based on the results of all the simulations.

Monte Carlo simulation is like playing a guessing game, where you try to estimate the outcome of a complicated situation by randomly guessing many times and then calculating the average of all your guesses. It's not just for predicting World Cup winners; it can be applied in finance, engineering, healthcare, transportation, gaming, energy, and environmental science. This can help you understand the probabilities of different outcomes in a complex situation, like a project deadline or a financial forecast. However, it's important to remember that these are just estimates and not exact calculations, and their success depends heavily on the accuracy of the parameters and models (thus the prediction can for sure be inaccurate as the picture captured above). Still, it can serve as an effective reference for the decision-making process if everything remains so uncertain.

That's all for this week's weekly digest! I hope you found the information on problem-solving methodologies and mindset from machine learning algorithms interesting and applicable to your daily life. As always, if you have any feedback or suggestions for future topics, please don't hesitate to reach out.

Until next time, stay curious and keep learning!

Best regards,

Sherman

Things that I found interesting this week:

📚 Book — “The Age of AI and Our Human Future” by Henry Kissinger, Eric Schimidt, and Daniel Huttenlocher

I recommend this book to you since I know you might be interested in the intersection of technology and society. The book is written by three influential authors, including Henry Kissinger, who is known to be the most influential figure in the field of international relationships. This book explores the potential of AI to impact our lives in ways we may not expect, including the use of AI in warfare and its potential implications for international relations. As AI becomes increasingly prevalent in warfare, questions arise about the role of governments in regulating AI development and use. For example, if a government encourages platforms to label or block certain content, or if it requires AI to identify and downgrade biased or “false” information, such decisions may effectively operate as an engine of social policy with unique breadth and influence. These types of policies could have significant implications for the future of warfare and international relations, and the book delves into the possibilities and challenges that arise with the use of AI in this context.

🎬  Video — The danger of predictive algorithms in criminal justice | Hany Farid | TEDxAmoskeagMillyard

I come across this really cool video on predictive algorithms in criminal justice by Hany Farid, a digital forensics expert and professor at UC Berkeley. The video talks about the inherent danger of predictive algorithms and the potential bias that comes with it. With the rise of AI and AGI, we can see the future where humans are giving more decision power to AI assuming they can be “more informed” and “rational” in some sense. But this video served as a great reminder of the importance of ethical considerations and human intervention in the decision-making process. TED talks are always a great source of inspiration and learning, and I recommend it to anyone interested in the intersection of AI and ethics.

📃 Blog — Monte Carlo Simulation: History, How it Works, and 4 Key Steps (investopedia.com)

I found this article pretty inspiring to me as it provides a more detailed introduction to the Monte Carlo simulation we previously discussed. The author not only explains the history and concept of Monte Carlo simulation but also provides practical steps on how to do it in Excel. The article is well-written and easy to understand, making it accessible to anyone who wants to learn about this powerful tool. I highly recommend this article to my fellow curious minds who want to explore the possibilities of Monte Carlo simulation.

Psst...hey there, did you find this week's digest interesting? If so, don't keep it to yourself! Sharing is caring, so why not forward it to your friends and let's spread the word together! Thanks for being an awesome friend!