Building Babylon · Scour

Gradient optimisation on the Poincaré disc

building-babylon.net·421w

Circle circumference in the hyperbolic plane is exponential in the radius: proof by computer game

building-babylon.net·421w

Hierarchical Softmax

building-babylon.net·457w

Minsky & Papert’s “Perceptrons”

building-babylon.net·464w

Re-parameterising for non-negativity yields multiplicative updates

building-babylon.net·471w

Factorisation of stochastic matrices

building-babylon.net·471w

Heider and Simmel misinterpreted

building-babylon.net·484w

An LCD digit dataset for illustrating the “parts-based” representation of NMF

building-babylon.net·487w

Don’t interpret linear hidden units, they don’t exist.

building-babylon.net·497w

Adagrad depends on the choice of co-ordinate system

building-babylon.net·499w

Orthogonal transformations and gradient updates

building-babylon.net·499w

Short-time Fourier transform cheatsheet

building-babylon.net·509w

The mathematics of the discrete Fourier transform

building-babylon.net·509w

Feature scaling and non-negative matrix factorisation

building-babylon.net·509w

Convergence rate of gradient descent

building-babylon.net·514w

Skipgram isn't Matrix Factorisation

building-babylon.net·520w

Softmax parameterisation and optimisation

building-babylon.net·523w

Improving Pairwise Learning for Item Recommendation from Implicit Feedback 2014

building-babylon.net·528w

WARP loss for implicit-feedback recommendation

building-babylon.net·528w

Parameterising the Mahalanobis distances for metric learning

building-babylon.net·547w

Log in to enable infinite scrolling