The adoption of large language models (LLMs) and vision-language models (VLMs) in edge AI applications has been limited by on-device computation and memory resources, and also by the cost of training and developing large models suitable for edge applications. The adoption of new training methods and algorithms combined with innovative new model architectures promises to dramatically change the outlook for adoption with models created for the edge that can efficiently execute within SWaP and cost targets.
The Impact of More Efficient Training and Inferencing of Large Language Models on Edge AI Applications
Session Speakers
-
Steven Brightfield
Chief Marketing Officer, BrainChip
Steven Brightfield has over 20 years of success defining and bringing to market new semiconductor products with companies such as Qualcomm Technologies, SiMA.ai, LSI Logic, Plessey Semiconductors and Zoran for the mobile, AR/VR, wearable, edge ML, cable/sat set-top and digital camera markets. He has 10 years of experience launching programmable semiconductor IP cores for CPU/GPU/DSP/NPUs at LSI Logic, ARC International PLC (acquired by Synopsys), MIPS, Siliconarts, Improv Systems and BOPS and licensing these for use in end products that are ubiquitous today. Earlier, Steven’s technical foundation in digital signal processing led to using DSPs in innovative products that digitized the world of speech, audio, multimedia, graphics, camera and video processing. Steven recently joined the BrainChip leadership team to further drive BrainChip’s brand recognition, go-to-market strategy and customer acquisition as BrainChip enters a growth phase with its flagship Akida products. Steven has a BS in Electrical Engineering from Purdue University.
