Compute-efficient deep learning: Algorithmic trends and opportunities
BR Bartoldson, B Kailkhura, D Blalock - Journal of Machine Learning …, 2023 - jmlr.org
Although deep learning has made great progress in recent years, the exploding economic
and environmental costs of training neural networks are becoming unsustainable. To …
and environmental costs of training neural networks are becoming unsustainable. To …
Beyond efficiency: A systematic survey of resource-efficient large language models
The burgeoning field of Large Language Models (LLMs), exemplified by sophisticated
models like OpenAI's ChatGPT, represents a significant advancement in artificial …
models like OpenAI's ChatGPT, represents a significant advancement in artificial …
EXACT: Scalable graph neural networks training via extreme activation compression
Training Graph Neural Networks (GNNs) on large graphs is a fundamental challenge due to
the high memory usage, which is mainly occupied by activations (eg, node embeddings) …
the high memory usage, which is mainly occupied by activations (eg, node embeddings) …
[PDF][PDF] Tinytrain: Deep neural network training at the extreme edge
On-device training is essential for user personalisation and privacy. With the pervasiveness
of IoT devices and microcontroller units (MCU), this task becomes more challenging due to …
of IoT devices and microcontroller units (MCU), this task becomes more challenging due to …
Back razor: Memory-efficient transfer learning by self-sparsified backpropagation
Transfer learning from the model trained on large datasets to customized downstream tasks
has been widely used as the pre-trained model can greatly boost the generalizability …
has been widely used as the pre-trained model can greatly boost the generalizability …
Tinykg: Memory-efficient training framework for knowledge graph neural recommender systems
There has been an explosion of interest in designing various Knowledge Graph Neural
Networks (KGNNs), which achieve state-of-the-art performance and provide great …
Networks (KGNNs), which achieve state-of-the-art performance and provide great …
TANGO: re-thinking quantization for graph neural network training on GPUs
Graph learning is becoming increasingly popular due to its superior performance in tackling
many grand challenges. While quantization is widely used to accelerate Graph Neural …
many grand challenges. While quantization is widely used to accelerate Graph Neural …
Fine-tuning language models over slow networks using activation quantization with guarantees
Communication compression is a crucial technique for modern distributed learning systems
to alleviate their communication bottlenecks over slower networks. Despite recent intensive …
to alleviate their communication bottlenecks over slower networks. Despite recent intensive …
ScheMoE: An Extensible Mixture-of-Experts Distributed Training System with Tasks Scheduling
In recent years, large-scale models can be easily scaled to trillions of parameters with
sparsely activated mixture-of-experts (MoE), which significantly improves the model quality …
sparsely activated mixture-of-experts (MoE), which significantly improves the model quality …
DIVISION: memory efficient training via dual activation precision
Activation compressed training provides a solution towards reducing the memory cost of
training deep neural networks (DNNs). However, state-of-the-art work combines a search of …
training deep neural networks (DNNs). However, state-of-the-art work combines a search of …