https://levilelis.github.io/notebook/tags/machine-learning/Recent content in Machine Learning on Research NotebookHugoen-usFri, 07 Aug 2026 00:00:00 +0000https://levilelis.github.io/notebook/posts/research/Fri, 07 Aug 2026 00:00:00 +0000https://levilelis.github.io/notebook/posts/research/<p>Current artificial agents require large amounts of data to learn to solve problems, and their solutions often fail to generalize to problems not seen during training. These weaknesses contrast with the learning abilities of animals, which can acquire complex skills that generalize to different scenarios from little experience. Animals are efficient learners in part because they rely on representations and learning mechanisms shaped by evolution, a process that is slow and computationally expensive. Our research attempts to close the gap between natural and artificial agents by studying how agents can learn representations that make future problems easier to solve. If the computational cost of intelligence is high and unavoidable, we can invest much of that computation in learning a reusable representation. Its cost can then be amortized across the many downstream problems that the representation makes easier to solve.</p>