Tag: statistics

On our final day, Danny and Sasha came back to show us the levels we created on our first day in the game. We got the chance to play through them and give them feedback on what we liked and what we thought could be improved.

Afterwards, we finished the alternative stats quiz. We changed the spatula to be the “flirtatious spatula”, and we came up with around six questions that had David Blackwell as the correct answer and two with the spatula as correct. 

I am extraordinarily thankful to Dr. Laber and everyone who came to teach us during the Work Experience Program for giving their time to help us grow and develop our interest in data science. Working at Laber Labs was incredibly rewarding and I was able to learn a lot throughout the program, and it furthered my decision to pursue a career in computer science in my post secondary education.

Since he had contracted COVID, Dr. Laber connected with us over Zoom today. We started the day off by learning how to plot a linear regression model in R. A linear regression model’s goal is to find the line of best fit on a scatter plot, and it does this by minimizing the loss (the distance between the points and the line of best fit on the y-axis). We also talked about K Nearest Neighbors, which is an algorithm that makes predictions about where a specific point may be. He also gave us the opportunity to write our own K Nearest Neighbors program where we write the same program he gave us but with multiple neighbors. This was my attempt to do so.

In the afternoon, we talked to Jesse Clifton, who’s a PhD student at NC State, about the ethics of Artificial Intelligence. Since machine learning is utilized by giving it a task, it’ll accomplish that task to the best of its ability by whatever means necessary. This means that if you give it a goal that isn’t perfect, it’ll find loopholes in order to maximize its rewards. The reason why it’s so important to have diversity in the field of artificial intelligence and machine learning is so that the goals of these algorithms aren’t determined by a group made up by people with the same viewpoint or perspective.

Justin Weltz, a PhD student at Duke, talked to us over a Zoom call about the pros and cons of random sampling as well as reinforcement learning, which is machine learning (which we touched upon with Alex and Dr. Laber) that concerns how agents should act in order to maximize the “rewards” they get (rewards are basically like treats for a program if it does what it’s told). He spoke about what fields he utilizes these in, such as precision medicine. Afterwards, we got the opportunity to speak with him about what being a PhD student is like and what doing research is like for him.

Today we talked to Alex Cloud, who founded Doran’s Lab and now works for Riot Games as a data scientist. 

Alex spoke about luck and skill used in games, using Randochess as an example. Randochess is a game in which you’ll first use a random number generator to determine whether you’ll flip a coin to see who wins or play chess. The idea here is to decide whether this game involves more luck or more skill if the numbers that correlate to flipping a coin is greater than the numbers that correlate to playing chess. Personally, I’d say that it still involves the same amount of skill, since if you’re better than chess than your opponent, you’d always have a better chance of winning.

Alex also showed us DALL-E 2, which creates art based on what the user puts in. The program was trained by feeding it a bunch of images with captions so that the program could start to identify patterns, so for example if you feed it two images with a dog in it, it’ll recognize that there are two similar animals in both images and both the captions mention a dog. However, it’s not completely perfect. An example of it not working as intended is if you feed it something like “tree bark”, to which it’ll give you an image of a dog barking at a tree rather than actual tree bark.

Dr. Laber gave a presentation on analyzing data and statistics to show us where it was used in the real world, such as the prime location to open a restaurant (and what factors to consider) and what data can tell us about where more armor should be placed on planes.

One the big examples he showed us was using machine learning to determine which emojis are the most used in sex trafficking cases. Since sex traffickers hide their advertisements through cryptic wording and emojis, it’s become increasingly hard to track them down, while still making their services accessible to their clients. Using machine learning, the program could identify whether someone was being trafficked/a minor or whether they were doing it out of their own will, as law enforcement prioritizes the former. We also talked about machine learning in precision medicine that determines the amount of dosage a patient should get by comparing the patient’s characteristics to those of former patients.