Deeplite AI - The Next XNOR.ai? Profile
Started by AI scientists, Deeplite AI is a Canadian analytics optimization company that claims to increase the processing speed and efficiency of analytics algorithms on the edge.
The closest comparable is XNOR.ai which took the market by surprise in 2019, empowering cheap Wyze cameras to beat Axis and Hikvision, shortly later terminating that deal and selling itself to Apple (and no longer available to 3rd parties). Could Deeplite deliver similar impact?
Based on an interview with the company, IPVM analyzes the company's background, positioning, and offerings.
**********
******** ** *** ******* ** * tech/SaaS ************,*** **** ******, *** * ******** *********** **.*.,*** ***** *******, ** ****. ******** ******** *** software ************ ******** *** ********* **** want ** ********** ***** ** ** make ** ****, ** ****. ********* to ********'* ************, **** **** * current ********* *** ** **** ***** projects *** *** ********* ****** *** USA, ******, ******, *** *****. ******** partners **********,********** **,********** ***,***, ********** ***.
*******
********* ** ********, ******** ** ************* *********** *******, *** ** ** ********* ****** $10 *******. ******** *** ********.**** * ********* ***** ****** ***** fundraising:
*** *** **** ** ******* ** investors ** *** ** ***** *****, kind ** ****** ********** ****.**, *** much **** **** **** ***.
****.** *** **** ** ******** ~$*** ******* ** ****.
*********
*** ******* ************ ** ****** *********,** ******* ** ******. ******** ** *** * ****** technical **** **** ** ********* *** scientists, ******* ** **** ********, ******* learning, *** **. ****** *** *********, Deeplite ****** ****** **** **% ** their ********* **** **.**.
*******
******** ****** ***** ******** *** ** optimization *** * *** ******* *** the ********* ******** ** ****** ********. Their **** ********* *** ********, **********, and ***********.
******** ********* ************* ********** ** "********* ***** ************ ********", where ***** ************** *** ****** ********* algorithms **** ****** (** *********), ******** (~x3 *********), *** ********* (*** *********) are ********** **** ******* ******** **** (<1.50%).
**'* **** **** ** ****** *** the ****** ***** *******. ** ** space ** ** *****, **** *** can **** ** ********™. ** ***, where ** **** *** **** ******* with **** ** **** **** (*************** units). ** **** *** **** ** the ****** *** *** *** ******** processors. *** **** ****, ** *** greatly ****** *** **** ** *** model *** ** **** ***** **** increasing *** *****.
******** **** ** **** ******** ******** results *** ******** **** ****** ** FP32 (******** ***** ** ****), ******* further ****** *** ****** ******, *** increased ***** ** ********* ***** ********* optimizations (**** ** ***** *********). ** fact, ********** **** ** *** ****** faster ** * ***** ********* ****** for ***-****, ***-***** *****.
**** *** **** * ***** **** FP32, *** **** *** ******** ** to ******* *****, ** *** ***** precision, ***. *** ** *******. ********** is *** *** ******* ****** *** ARM ****** * *** ** ********* to *** ***. *** **** ** very **** ******* ** *** **** technical **** ** *** ** ******** these *****.
******** ********* *********** ** "*** ****** models ** ******* *** *** *********, and ** ******* ***** ** *** customers."
*******, ******** ********* ********'* *** ********* annotation ****:
** * ******** *** **** ***** open ****** **** *** ****'** ****** certain ******** **** ******** ** ******* sites, **** ** ***** ** **, basically **** **** ****** **** **** site, **** **** ** *** ***** and *** * ***** ***** ** auto ***** **** ****. **** **** those ********* ****** *** ******* * small ***** ** *** **** ******, and ****** ** ** **** **** where ***** *** ******** *** *** so **** ** **** ******* **** accuracy.
******** ******** *******, ********** **** *** Onyx, *** **** ** ****** ** C++.
*******
******** ***** **** *** ****** ******* after *****'* *********** ** ****. *******, we *** *** **********: (*) ******** to * ***** ***, **** ********* in **** ******** *** ******* ***** with ********** ********* *** ***** **** powerful ******** *** (*) **** ***** want ** ** ** ********** *** not ** ********* ** *** ******* (especially *** ****** ****). *******, *** potential ********* ******** ******* ***** ******* analytics ** ***** **** ******** ** more ******** ********* **** ************* ***** otherwise ***** ** *****.
**** ** ******* ********'* *********** *** comparing ** ** ***-***-***** ******** ** the **********.