Chiplets for addressing Challenges and Use cases for AI and ML applications.pdf

1、Why,how and what of chiplets for AI/ML spaceDharmesh Jani(“DJ”)Infra Ecosystem and Partnership LeadChiplet use cases in AI and ML ChipletsChiplet Use cases in AI/MLDharmesh Jani(“DJ”)Infra Ecosystem and Partnership LeadMetaWhat are universal drivers?Universal Human Use CasesArc of the talkWhat are u

2、niversal drivers?Universal Human Use CasesAI/ML as proven pathSystem DesignDSA ChallengesChiplets applicationsWhy is AI/ML the path forward?Implications of AI SOTA workHow chiplets can play a role for the industry here?Arc of the talkRecognitionMiningSynthesisUniversal use cases that are drive techn

3、ologyFundamental use cases have recurring theme of recognition,mining and synthesis for learning and knowledge creationRecognitionBuild identification models by machines of real worldRecognition is the“what is”and create a canonical representative modelRequires training!Universal use cases that are

4、drive technologyMiningSearch instances of the model in the sea of dataMining is searching across all forms of data(e.g.,Image,text,video,logs etc.)Requires inference!Universal use cases that are drive technologySynthesisCreating new instance of models where one does not existSynthesis is creation by

5、 machines of new ideas Requires multi-modality,GANs!Universal use cases that are drive technologyWhat are universal drivers?Universal Human Use CasesAI/ML as proven pathWhy is AI/ML the path forward?Arch of the talk15011001150120012501300186 87 88 89 90 91 92 93 94 95 96 97 98 99 00 01 02 03 04 05 0

6、6 07 08 09 10 11 12 13 14 15 16 17IncreaseYearGrowth of the term deep learning in researchSource:MIT Technology ReviewWhat Happened Here?15011001150120012501300186 87 88 89 90 91 92 93 94 95 96 97 98 99 00 01 02 03 04 05 06 07 08 09 10 11 12 13 14 15 16 17IncreaseYearGrowth of the term deep learning