English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Tools annual

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Tools monthly

Global: Zendy Plus monthly

Presents the access of premium content as premium feature

Premium Content

Global: Zendy Plus annual

Global: Zendy Free Plan

Training deep learning models takes an extremely long execution time andconsumes large amounts of computing resources. At the same time, recentresearch proposed systems and compilers that are expected to decrease deeplearning models runtime. An effective optimisation methodology in dataprocessing is desirable, and the reduction of compute requirements of deeplearning models is the focus of extensive research.  In this paper, we address the neural network sub-graph transformation byexploring reinforcement learning (RL) agents to achieve performanceimprovement. Our proposed approach RLFlow can learn to perform neural networksubgraph transformations, without the need for expertly designed heuristics toachieve a high level of performance.  Recent work has aimed at applying RL to computer systems with some success,especially using model-free RL techniques. Model-based reinforcement learningmethods have seen an increased focus in research as they can be used to learnthe transition dynamics of the environment; this can be leveraged to train anagent using a hallucinogenic environment such as World Model (WM), therebyincreasing sample efficiency compared to model-free approaches. WM usesvariational auto-encoders and it builds a model of the system and allowsexploring the model in an inexpensive way.  In RLFlow, we propose a design for a model-based agent with WM which learnsto optimise the architecture of neural networks by performing a sequence ofsub-graph transformations to reduce model runtime. We show that our approachcan match the state-of-the-art performance on common convolutional networks andoutperforms by up to 5% those based on transformer-style architectures

RLFlow: Optimising Neural Network Subgraph Transformation with World Models