High-dimensional SGD aligns with emerging outlier eigenspaces.
Gérard Ben ArousReza GheissariJiaoyang HuangAukosh JagannathPublished in: ICLR (2024)
Keyphrases
- high dimensional
- nearest neighbor search
- outlier detection
- dimensionality reduction
- low dimensional
- sparse data
- multi dimensional
- high dimensionality
- similarity search
- high dimensional data
- nearest neighbor
- variable selection
- input space
- data points
- feature space
- high dimension
- metric space
- parameter space
- stochastic gradient descent
- high dimensional problems
- neural network
- training samples
- multi modal
- similarity measure
- machine learning