GANs, diffusion models, and the architectures behind modern generative AI. Research-grade explainers on how generative models learn, where they fail, and the design choices that separate state-of-the-art methods from the rest, each tied to the paper it came from.
A photo restoration shop receives a folder of pictures taken on a ten year old phone and runs them through a super resolution model with excellent benchmark scores. The results look worse than the originals. Grain turns into worms, text…
Real World Super Resolution With Rectified Flow Degradation Read More »
Most work on generating dynamic 3D scenes treats motion as something to reconstruct, not something to control. A neural network watches a video, learns a warp field that explains how every point moved across every frame, and the result plays…
SkeletonGaussian Rigs Video Into Editable 4D Animation Read More »
A researcher trains a GAN on faces. For two hours the samples improve. Then, over about ten minutes, every sample turns into the same smiling woman with slightly different hair. The losses never exploded. Nothing crashed. The generator simply found…
GAN Mathematics, Divergences, Duality and Training Dynamics Read More »
Start with the biology, because the whole idea rests on it. A T cell patrols the body carrying a receptor, its own personal sensor, tuned to recognize short protein fragments called peptides. Those peptides are held up for inspection by…
Tcr-Translate: An AI That Translates Targets Into T Cell Receptors Read More »
Designing a protein from scratch on a computer is now routine for small ones. Ask for a large protein, past about 400 amino acids, and the leading generators start to buckle, both because their math scales badly and because each…
Salad Designs Large Proteins Fast With Sparse Denoising Read More »
Teaching a computer to carve an image into its objects, every person, every car, every patch of sky, has always meant paying people to trace those outlines by hand. The best tool for the job, Segment Anything, was trained on…
Diffuse2Seg: How Diffusion Models Segment Images Without Labels Read More »
AlphaFold changed biology by cracking a puzzle that had resisted everyone for fifty years. Hand it a protein’s sequence and it tells you the shape that sequence folds into. The answer it gives is a single, crisp structure, and that…
Mac-Diff: A Diffusion Model for Diverse Protein Conformations Read More »
Ask a generative model to dream up a new crystal and it will usually hand you something that looks a lot like what it already knows. That is not a flaw. It was trained to reproduce the patterns in its…
Reinforcement Learning Guides Diffusion to New Crystals Read More »
Start with the scale of the problem, because it is what makes this work matter. Bacteria evolve, and every time we deploy an antibiotic widely, we hand them a reason to develop resistance to it. Over decades this has hollowed…
ApexGO Optimizes Peptide Antibiotics With Generative AI Read More »
Diffusion models got very good very fast at turning text into images. What they did not automatically get good at was matching what a person actually wants to see. A prompt asking for a cat sleeping in a sunbeam might…
Why DPO Is Beating RLHF at Aligning AI Images Read More »