PPO has a reputation for being the “stable” reinforcement learning algorithm. That reputation is mostly deserved. Compared with older policy-gradient methods, PPO is easier to get running, easier to debug, and less likely to destroy a policy with one bad update. But “stable” is not the same thing as “insensitive.” In a previous experiment, How Sensitive […]

Artificial Intelligence

How KNN Actually Works

K-Nearest Neighbors, or KNN, is one of the most intuitive supervised learning methods: to predict a new point, it finds the k training examples that are closest under a chosen distance metric, then uses those neighbors to make the prediction. In classification, the prediction is usually a majority vote of nearby labels; in regression, it […]

Artificial Intelligence

I Profiled Bad GEMM Kernels. Shared Memory Wasn’t the First Win.

I broke CUDA matrix multiplication on purpose, fixed one bottleneck at a time, and measured which optimizations actually moved performance. CUDA optimization advice is everywhere: use shared memory, improve occupancy, coalesce memory, unroll loops, reduce synchronization, avoid bank conflicts. All of that advice can be true, but it is not equally important at every stage. […]

Explore AI articles & research

On YouTube