Energy-efficiency Limits on Training AI Systems using Learning-in-Memory

arXiv:2402.14878v1 Announce Type: new
Abstract: Learning-in-memory (LIM) is a recently proposed paradigm to overcome fundamental memory bottlenecks in training machine learning systems. While compute-in-memory (CIM) approaches can address the so-called memory-wall (i.e. energy dissipated due to repeated memory read access) they are agnostic to the energy dissipated due to repeated memory writes at the precision required for training (the update-wall), and they don't account for the energy dissipated when transferring information between short-term and long-term memories (the consolidation-wall). The LIM …

abstract ai systems arxiv bottlenecks compute cs.ai cs.ar cs.lg efficiency energy in-memory learning systems machine machine learning memory paradigm systems training training ai type