AI LabAI Lab

2022

7/

27

Edge AI Evangelist’s Thoughts Vol.22: Comparing the Carbon Footprint of Sparse Modeling and Deep Learning

ThisisHaruyukiTago,EdgeEvangelistatHACARUS’TokyoR&Dcenter.Inthisseriesofarticles,IwillsharesomeinsightsfrommydecadesofexperienceinthesemiconductorindustryandcommentonvariousAIindustry-relatedtopicsfrommyuniqueperspective.Intoday’svolume,Iwillexplainalittleaboutsparsemodelinganddeeplearningmethodsandcomparetheircarbonfootprints.Recently,SDGshavebecomeahottopicintoday'sbusinessenvironment,especiallyintheAIindustry.Whenlookingat…

2022

6/

27

Edge AI Evangelist’s Thoughts Vol.21: ARM vs RISC-V

Helloeveryone,thisisHaruyukiTago,EdgeEvangelistatHACARUS’TokyoR&Dcenter.Inthisseriesofarticles,IwillsharesomeinsightsfrommydecadesofexperienceinthesemiconductorindustryandIwillcommentonvariousAIindustry-relatedtopicsfrommyuniqueperspective.Intoday’svolume,Iwanttodiveintothemicroprocessorindustryoflogicintegratedcircuitsbeingusedincellphones,embeddeddevices,digitalhomeappliances,andotherbusinesses.Below,wewilldiscussthebusinessof…

2022

6/

10

Edge AI Evangelist’s Thoughts Vol.20: The Secrets of Apple’s M1 Ultra

Helloeveryone,thisisHaruyukiTago,EdgeEvangelistatHACARUS’TokyoR&Dcenter.Inthisseriesofarticles,IwillsharesomeinsightsfrommydecadesofexperienceinthesemiconductorindustryandIwillcommentonvariousAIindustry-relatedtopicsfrommyuniqueperspective.Intoday’svolume,IwanttopresentmyfindingsonApple’sM1UltramicroprocessorandMacStudioPC,whichwereannouncedinMarchof2022.Inthisarticle,Iwillcoverthespecificationsandperformance…

2022

4/

06

Utilizing the Icecream Package for Efficient Print Debugging & How to Customize It

Hello!ThisisAtsushiYasumoto,adatascientistatHACARUS. Inthisarticle,Iwillgiveanoverviewoftheicecreampackage,ausefultoolfordebuggingprintwhileusingPython.IwillalsointroduceacustomizedfeaturewehaveaddedatHACARUSaspartofourwork.Beforetryingthisforyourself,pleaseusepipinstallicecreamtoinstallthev2.1.2versionoficecream. Tostart,let’slookataquickexampleofprintdebugging.Supposeyouhavean…

2022

4/

05

Edge AI Evangelist’s Thoughts Vol.19:  Analyzing Unconventional Logic Semiconductors – A Shift Away from Semiconductor Manufacturers

Helloeveryone,thisisHaruyukiTago,EdgeEvangelistatHACARUS'TokyoR&Dcenter.Inthisseriesofarticles,IwillsharesomeinsightsfrommydecadesofexperienceinthesemiconductorindustryandIwillcommentonvariousAIindustry-relatedtopicsfrommyuniqueperspective.Intoday’svolume,Iwillonceagaincoverthreemaintopics.First,IwilldiscusstheestimatednumberofserversusedbyAmazon,theworld’slargestonlineretailer,andwhyin-housedevelopmentisadvantageousfor…

2022

3/

23

Edge AI Evangelist’s Thoughts Vol.18: Decarbonizing AI development by a factor of 100 compared to Deep Learning AI

Helloeveryone,thisisHaruyukiTago,EdgeEvangelistatHACARUS'TokyoR&Dcenter.Inthisseriesofarticles,IwillsharesomeinsightsfrommydecadesofexperienceinthesemiconductorindustryandIwillcommentonvariousAIindustry-relatedtopicsfrommyuniqueperspective.Intoday’svolume,Iwilltrymybesttocoverthreetopics.Thefirstinvolvesexaminingthetransitionofcomputationalpowerduringmachinelearningtraining.ThesecondexploreswhythepowerconsumptionandCO2emissions…

2022

2/

07

Edge AI Evangelist’s Thoughts Vol.17: Measuring a Bluetooth Microcontroller’s Battery Life

Helloeveryone,thisisHaruyukiTago,EdgeEvangelistatHACARUS'TokyoR&Dcenter.Inthisseriesofarticles,IwillsharesomeinsightsfrommydecadesofexperienceinthesemiconductorindustryandIwillcommentonvariousAIindustry-relatedtopicsfrommyuniqueperspective.BatteryLifeModel&PredictionsforBluetoothMicrocontrollers Previouslyinvolume15oftheEdgeAIEvangelistThoughtsseries[1],IexploredusingaBluetoothLow-powerThermometerprogramusingthe‘Thunderboard’bySiliconlabs[2].…

2022

1/

14

Edge AI Evangelist’s Thoughts Vol.16: Smart Edge Devices with Spresense

Helloeveryone,thisisHaruyukiTago,EdgeEvangelistatHACARUS'TokyoR&Dcenter.Inthisseriesofarticles,IwillsharesomeinsightsfrommydecadesofexperienceinthesemiconductorindustryandIwillcommentonvariousAIindustry-relatedtopicsfrommyuniqueperspective.Inthisvolume,IwilltalkaboutarecentexperimentIconductedusingSony’sSpresenseboardtomonitortherotationalnoisefromamodelmotor.TheSpresensewasalsousedtodetermineifthemotor…

2022

1/

05

RuleFit: A Modeling Method for Automatically Extracting Interactions

IntroductionHelloeveryone,thankyouforreadingmyarticletoday.MynameisYushiro,adatascientistatHACARUS.AtHACARUS,weworkwithvariousmachinelearningmodelsandwealsowriteblogsaboutlearningandinterpretationmethodsforAI. Inmypreviousarticle,IintroducedthetopicofPDP,ICEplots,andd-ICE.Theseareallinterpretationmethodsformachinelearningmodelsthatcanbeappliedtoanymodel.WhilePDPandICEplotsareusefulforvisualizingthe…

2021

12/

28

Visualizing AI: PDP Reliability using d-ICE

IntroductionHelloeveryone,thisisYushiro,adatascientistatHACARUS. Thesedays,withlibrariessuchasScikit-learn,TensorFlow,andPyTorch,itisbecomingincreasinglyeasytocreatemachinelearningmodelsusingdata.Thesemodelsarecommonlyusedtomakepredictionsfornewdatabasedondatafromthepast. However,asthesemodelsbecomemorecomplex,theissuesofblack-boxingbecomemoreprevalent.Thismakesitdifficultforuserstounderstandthereasoningbehindthemodel’sdecision-making.Black-boxingis…