Enhanced regularization for on-chip training using analog and temporary memory weights.
Raghav SinghalVivek SaraswatShreyas DeshmukhSreenivas SubramoneyLaxmeesha SomappaMaryam Shojaei BaghiniUdayan GangulyPublished in: Neural Networks (2023)
Keyphrases
- analog vlsi
- circuit design
- early stopping
- low cost
- feedforward artificial neural networks
- high speed
- memory subsystem
- training set
- mixed signal
- high density
- supervised learning
- linearly combined
- processor core
- weighted sum
- training process
- linear combination
- memory access
- multi channel
- memory requirements
- memory space
- stochastic gradient descent
- restricted boltzmann machine
- focal plane
- regularization parameter
- training examples