PinnedInTowards AIbyArjun Bhasin·Oct 25How SVD Compresses an Image — One Layer at a TimeEvery photo can be rebuilt from simple layers, and the first few carry most of the picture.A response icon1A response icon1
PinnedInAI Advances BriefsbyArjun Bhasin·Sep 24The Kelly Criterion for AI AgentsWhy a Better Model Doesn’t Deserve More PowerA response icon1A response icon1
InTowards AIbyArjun Bhasin·Nov 6The Fourier Transform Is Just a Change of BasisHow the same signal gets one set of coordinates in time and another in waves.A response icon1A response icon1
InTowards AIbyArjun Bhasin·Nov 3How Can Two Matrices of Different Sizes Have the Same Eigenvalues?Why a large eigenvalue problem can be solved through a much smaller matrix.A response icon1A response icon1
InTowards AIbyArjun Bhasin·Oct 28How a Matrix Trick Helped Win a Million-Dollar PrizeHow Netflix guesses what you will watch. And why a huge table of ratings hides a few simple ‘tastes’.
InTowards AIbyArjun Bhasin·Oct 26How to Invert a Matrix That Has No InverseHow the pseudo-inverse helps when a system has too many solutions, or none at all.A response icon1A response icon1
InTowards AIbyArjun Bhasin·Oct 22Why Changing the Order of 3D Rotations Changes the ResultTwo 90° turns can leave an object facing two different ways, depending only on which turn comes first.A response icon2A response icon2
InPython in Plain EnglishbyArjun Bhasin·Oct 22Understanding Kinematic Control of Planar ManipulatorsHow a robot arm turns joint angles into a hand position, and how it finds its way back.A response icon1A response icon1
InTowards AIbyArjun Bhasin·Oct 20Why Your Model Blurs Even When It’s RightThe best squared-error prediction is an average, and that average is a projection.A response icon2A response icon2
InTowards AIbyArjun Bhasin·Oct 18Can You Tell Which Matrix Will Explode?Two matrices differ by a single minus sign. A small error in the data barely changes one answer, but changes the other by a lot.A response icon1A response icon1
![For A = [3 4], the shortest exact solution x_parallel = (0.6, 0.8) lies in the row space. Adding the perpendicular null-space vector z = (0.8, -0.6) gives x = (1.4, 0.2). Both inputs produce 5, but the squared length increases from 1 to 2. A right-angle marker shows the orthogonal decomposition.](https://miro.medium.com/v2/resize:fill:160:106/1*2Q4ELIZJVaSXbwchvHRkBQ.png)
![For A = [3 4], the shortest exact solution x_parallel = (0.6, 0.8) lies in the row space. Adding the perpendicular null-space vector z = (0.8, -0.6) gives x = (1.4, 0.2). Both inputs produce 5, but the squared length increases from 1 to 2. A right-angle marker shows the orthogonal decomposition.](https://miro.medium.com/v2/resize:fill:320:214/1*2Q4ELIZJVaSXbwchvHRkBQ.png)
