1 minute read
Works Machine Learning
from Naturally Stupid
by LASA Ezine
Roberto Martin-Martin is an Assistant Professor of Computer Science at the University of Texas and is researching AI and robotics-related topics. He studied Microelectronics and Electrical Engineering and started to do research in Robotics in Germany, and then later got a Ph.D. in robotics. Martin-Martin went into the field of AI because he “was always interested to know how machines work, how they make decisions on their own and how can we make them better at that based on sensor input.”
“I’m working on reinforcement learning algorithms to learn to move the entire body of a robot that can navigate and use its arms to interact. My goal is to create a robot that can help with household chores.”
Advertisement
- Roberto Martin-Martin
There are three main groups of AI: machine learning, neural networks, and deep learning. Machine Learning refers to algorithms that are used to gradually improve performance. Martin-Martin says that there are many different types of machine learning, like supervised learning and unsupervised learning. Supervised learning, he states, “[assumes] we have a dataset of labeled cases.” In the words of DJ Franklin, “In order to train a model to identify dogs in an image, we’d have to feed lots of images with dogs and images without dogs into the model so it can learn to distinguish between them.” They use patterns to understand the data set they are given.
Neural Networks, a smaller group within machine learning, imitate the way human brains work. You get an input, your brain does something to that input, and then you return the output. In Neural Networks, you have the input layer, the hidden layer (which modifies the input), and then the output layer. Last but not least, deep learning is just a type of Neural Network that has more than one hidden layer. For many AI models, people have tested and researched them enough to completely understand their capabilities and their limits. However, practically every new AI is improving on its previous version, and AI is becoming more and more powerful. Newer models though, like Chat GPT are still not yet fully understood because of their recency and their complexity. Newer language models have billions of parameters they take into account when producing an output.