In this piece, I will be writing about Loss Change Allocation (LCA), a novel method introduced by Uber AI Labs that helps provide visibility into the training process of deep neural networks.

In a talk at Samsung Next, Jason Yosinski (an author of the LCA paper) illustrates the need for such techniques by plotting the rise of compute and increased access to data against our scientific understanding of machine learning models. Yosinski sees this widening gap between our capabilities and our understanding as an opportunity for a new field, something akin to “AI Neuroscience”. …


A Deep Q-Learning approach to solving a common puzzle

Source: https://openai.com/blog/solving-rubiks-cube/. OpenAI made headlines in Fall 2019 for solving the related but more difficult RL task of teaching a physical hand to manipulate the cube. The task of actually solving the cube was done using Kociemba’s Algorithm.

Welcome to the second installment of my attempt to solve a Rubik’s Cube via reinforcement learning (RL). Last time, I provided an intro to Markov Decision Processes (MDPs) and formulated the task of solving a Rubik’s Cube as an MDP. If you missed this post or would like a quick refresher, you can check it out here.

At the end of my last post, I left off with a discussion of the Q-function and how we will need to approximate it for our task since the space of state-action pairs is too large. In this post, I will implement a…


A Deep Q-Learning approach to solving a common puzzle

Source: https://openai.com/blog/solving-rubiks-cube/. OpenAI made headlines in Fall 2019 for solving the related but more difficult RL task of teaching a physical hand to manipulate the cube. The task of actually solving the cube was done using Kociemba’s Algorithm.

Last year, I started my journey into machine learning through a Master’s program at Cornell Tech. One topic that particularly caught my eye was reinforcement learning (RL), which we approached from both the traditional direction of Markov Decision Processes (MDP) and from the direction of Deep Learning (DL). While the coursework was very informative, I wanted to take it a step further. Here, I have documented my (ongoing) attempt to do just that, by training an agent to solve a Rubik’s Cube.

A couple introductory notes:

  1. Solving a Rubik’s Cube with reinforcement learning is not a new problem and I…

Matthew Dalton

Data Scientist, M.Eng. in Operations Research ‘19

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store