27:
217:. By developing smaller DNNs, the firm has been able to run deep learning on scaled-down processing hardware such as smartphones and automotive-grade chips. In 2018, the firm said that its engineering team had moved beyond SqueezeNet and that it had developed even faster and more accurate DNNs for use in commercial products.
265:
In
January 2019, the firm launched an automotive perception software product called "Carver" that uses deep neural networks to perform object detection, lane identification, and drivable area identification. To accomplish this, Carver uses three neural networks which run in parallel. While running in
233:
NAS to design a family of fast and accurate DNNs for semantic segmentation of images. The paper claimed that the
SqueezeNAS neural networks outperform the speed-accuracy tradeoff curve of Google's MobileNetV3 family of neural network models. While Google used thousands of GPU-days to search for the
741:
Howard, Andrew; Sandler, Mark; Chu, Grace; Chen, Liang-Chieh; Chen, Bo; Tan, Mingxing; Wang, Weijun; Zhu, Yukun; Pang, Ruoming; Vasudevan, Vijay; Le, Quoc V.; Adam, Hartwig (2019-05-06). "Searching for MobileNetV3".
242:
The firm develops perceptual system software which uses deep neural networks to enable cars to interpret their environment. The software is designed for integration into an open platform, where a wide range of
569:
Iandola, Forrest N; Han, Song; Moskewicz, Matthew W; Ashraf, Khalid; Dally, William J; Keutzer, Kurt (2016). "SqueezeNet: AlexNet-level accuracy with 50x fewer parameters and <0.5MB model size".
888:
873:
229:(NAS) has begun to outperform humans at designing DNNs that produce high-accuracy results while running fast. In 2019, DeepScale published a paper called SqueezeNAS, which used
883:
599:
266:
real-time, these three networks perform a total of 0.6 trillion operations per second ("tera-ops/sec"). As a point of reference, each of the two redundant chips on the
789:
878:
847:
893:
720:
Shaw, Albert; Hunter, Daniel; Iandola, Forrest; Sidhu, Sammy (2019). "SqueezeNAS: Fast neural architecture search for faster semantic segmentation".
699:
Zoph, Barret; Vasudevan, Vijay; Shlens, Jonathon; Le, Quoc V. (2017-07-21). "Learning
Transferable Architectures for Scalable Image Recognition".
814:
371:
346:
202:
234:
design of MobileNetV3, DeepScale used just tens of GPU-days to automatically design the DNNs presented in the
SqueezeNAS paper.
591:
209:(DNNs) more efficient. In 2016, shortly after the founding of DeepScale, Iandola, Keutzer, and their collaborators released
839:
230:
226:
621:
143:
102:
248:
321:
206:
545:
442:
412:
743:
721:
700:
570:
391:
290:
270:
can perform 36 tera-ops/sec. So 0.6 tera-ops/sec is only 2% of the capacity of each Tesla chip.
58:
790:"DeepScale Announces Carver21: Modular Deep Learning Perception Software for Driver-Assistance"
649:
347:"Tesla is buying computer vision start-up DeepScale in a quest to create truly driverless cars"
151:
147:
259:
178:
650:"Why Tesla Quietly Acquired DeepScale, a Machine Learning Startup That's 'Squeezing' A.I."
295:
267:
214:
186:
181:. In 2018, the firm announced strategic partnerships with automotive suppliers including
166:
517:
201:
Prior to the founding of DeepScale, Forrest
Iandola and Kurt Keutzer worked together at
294:
stated that "it's apparent that DeepScale's technology will be integrated into Tesla's
764:
467:
867:
840:"Tesla reportedly buys machine-learning startup DeepScale for self-driving car tech"
251:
can be used. The software is able to run on a variety of processors, ranging from
170:
63:
372:"SqueezeBERT promises faster mobile NLP while maintaining BERT levels of accuracy"
262:-based processing chips that are designed specifically for the automotive market.
376:
190:
155:
92:
675:
304:
reported that "DeepScale's approach to autonomy fits the bigger picture that
285:
210:
298:, the self-driving technology the company is currently working on." Further,
492:
305:
26:
396:
815:"Tesla's new self-driving chip is here, and this is your best look yet"
182:
313:
252:
244:
748:
726:
705:
575:
317:
309:
676:"How to become a Full-Stack Deep Learning Engineer (time: 51:30)"
300:
280:
255:
324:
sensors will make up a robust system without other hardware."
116:
592:"Tesla Beefs Up Autonomy Effort With DeepScale Acqui-Hire"
308:
has promoted for a few years now. Rather than relying on
128:
142:
was an
American technology company headquartered in
123:
109:
98:
87:
72:
51:
33:
154:. On October 1, 2019, the company was acquired by
189:. On October 1, 2019, the firm was purchased by
213:, which is a small and energy-efficient DNN for
193:, which works on autonomous vehicle technology.
889:Defunct software companies of the United States
622:"Steve Cohen Buys The Dip In Self-Driving Cars"
268:Tesla Full Self-Driving computer system board
8:
874:Companies based in Mountain View, California
19:
443:"Visteon Works with DNN Vanguard DeepScale"
25:
18:
884:Technology companies of the United States
747:
725:
704:
574:
468:"Are We Short of Deep Learning Experts?"
392:"Are We Short of Deep Learning Experts?"
340:
338:
336:
332:
879:Software companies based in California
546:"DeepScale on Robo-Car: Fuse Raw Data"
765:"Does Your AI Chip Have Its Own DNN?"
670:
668:
643:
641:
7:
539:
537:
436:
434:
432:
165:DeepScale was co-founded in 2015 by
680:Silicon Valley Deep Learning Group
590:Niedermeyer, Edward (2019-10-01).
203:University of California, Berkeley
14:
894:2015 establishments in California
312:, Musk has consistently believed
850:from the original on 2019-10-10
838:Szymkowski, Sean (2019-10-02).
602:from the original on 2019-10-02
411:Marinova, Polina (2018-04-04).
187:Hella Aglaia Mobile Vision GmbH
813:Hollister, Sean (2019-04-22).
1:
763:Yoshida, Junko (2019-08-25).
648:Reisinger, Don (2019-10-02).
544:Yoshida, Junko (2017-09-21).
466:Yoshida, Junko (2018-04-03).
441:Yoshida, Junko (2018-01-09).
370:Johnson, Khari (2020-06-23).
345:Kolodny, Lora (2019-10-01).
173:. In 2018, DeepScale raised
16:American technology company
910:
788:Landen, Ben (2019-01-25).
620:Shazar, Jon (2018-04-05).
227:neural architecture search
221:Neural architecture search
39:; 9 years ago
288:had purchased DeepScale.
144:Mountain View, California
103:Mountain View, California
76:October 1, 2019
24:
278:On October 1, 2019,
274:Acquisition by Tesla
207:deep neural networks
115:Forrest N. Iandola (
21:
231:supernetwork-based
152:automated vehicles
59:Forrest N. Iandola
493:"Faculty Webpage"
225:In recent years,
150:technologies for
148:perceptual system
146:, that developed
137:
136:
901:
859:
858:
856:
855:
835:
829:
828:
826:
825:
810:
804:
803:
801:
800:
785:
779:
778:
776:
775:
760:
754:
753:
751:
738:
732:
731:
729:
717:
711:
710:
708:
696:
690:
689:
687:
686:
672:
663:
662:
660:
659:
645:
636:
635:
633:
632:
617:
611:
610:
608:
607:
587:
581:
580:
578:
566:
560:
559:
557:
556:
541:
532:
531:
529:
528:
513:
507:
506:
504:
503:
488:
482:
481:
479:
478:
463:
457:
456:
454:
453:
438:
427:
426:
424:
423:
408:
402:
401:
388:
382:
381:
367:
361:
360:
358:
357:
342:
179:Series A funding
176:
133:
130:
83:
81:
47:
45:
40:
29:
22:
909:
908:
904:
903:
902:
900:
899:
898:
864:
863:
862:
853:
851:
837:
836:
832:
823:
821:
812:
811:
807:
798:
796:
787:
786:
782:
773:
771:
762:
761:
757:
740:
739:
735:
719:
718:
714:
698:
697:
693:
684:
682:
674:
673:
666:
657:
655:
647:
646:
639:
630:
628:
619:
618:
614:
605:
603:
589:
588:
584:
568:
567:
563:
554:
552:
543:
542:
535:
526:
524:
516:Keutzer, Kurt.
515:
514:
510:
501:
499:
491:Keutzer, Kurt.
490:
489:
485:
476:
474:
465:
464:
460:
451:
449:
440:
439:
430:
421:
419:
410:
409:
405:
390:
389:
385:
369:
368:
364:
355:
353:
344:
343:
334:
330:
276:
240:
223:
215:computer vision
199:
174:
167:Forrest Iandola
163:
140:DeepScale, Inc.
127:
112:
79:
77:
68:
43:
41:
38:
20:DeepScale, Inc.
17:
12:
11:
5:
907:
905:
897:
896:
891:
886:
881:
876:
866:
865:
861:
860:
830:
805:
794:DeepScale Blog
780:
755:
733:
712:
691:
664:
637:
612:
582:
561:
533:
508:
483:
458:
428:
403:
383:
362:
331:
329:
326:
284:reported that
275:
272:
239:
236:
222:
219:
198:
195:
175:US$ 15 million
162:
159:
135:
134:
125:
121:
120:
113:
110:
107:
106:
100:
96:
95:
89:
85:
84:
74:
70:
69:
67:
66:
61:
55:
53:
49:
48:
35:
31:
30:
15:
13:
10:
9:
6:
4:
3:
2:
906:
895:
892:
890:
887:
885:
882:
880:
877:
875:
872:
871:
869:
849:
845:
841:
834:
831:
820:
816:
809:
806:
795:
791:
784:
781:
770:
766:
759:
756:
750:
745:
737:
734:
728:
723:
716:
713:
707:
702:
695:
692:
681:
677:
671:
669:
665:
654:
651:
644:
642:
638:
627:
623:
616:
613:
601:
597:
593:
586:
583:
577:
572:
565:
562:
551:
547:
540:
538:
534:
523:
519:
512:
509:
498:
494:
487:
484:
473:
469:
462:
459:
448:
444:
437:
435:
433:
429:
418:
414:
407:
404:
400:. 2018-04-03.
399:
398:
393:
387:
384:
379:
378:
373:
366:
363:
352:
348:
341:
339:
337:
333:
327:
325:
323:
319:
315:
311:
307:
303:
302:
297:
293:
292:
287:
283:
282:
273:
271:
269:
263:
261:
257:
254:
250:
246:
237:
235:
232:
228:
220:
218:
216:
212:
208:
204:
196:
194:
192:
188:
184:
180:
172:
168:
160:
158:
157:
153:
149:
145:
141:
132:
126:
122:
118:
114:
108:
104:
101:
97:
94:
90:
86:
75:
71:
65:
62:
60:
57:
56:
54:
50:
36:
32:
28:
23:
852:. Retrieved
843:
833:
822:. Retrieved
818:
808:
797:. Retrieved
793:
783:
772:. Retrieved
768:
758:
736:
715:
694:
683:. Retrieved
679:
656:. Retrieved
652:
629:. Retrieved
625:
615:
604:. Retrieved
595:
585:
564:
553:. Retrieved
549:
525:. Retrieved
521:
511:
500:. Retrieved
496:
486:
475:. Retrieved
471:
461:
450:. Retrieved
446:
420:. Retrieved
416:
413:"Term Sheet"
406:
395:
386:
375:
365:
354:. Retrieved
350:
299:
289:
279:
277:
264:
241:
224:
205:, on making
200:
171:Kurt Keutzer
164:
139:
138:
99:Headquarters
91:Acquired by
64:Kurt Keutzer
626:Dealbreaker
522:UC Berkeley
497:UC Berkeley
377:VentureBeat
258:to smaller
156:Tesla, Inc.
93:Tesla, Inc.
868:Categories
854:2019-11-10
824:2020-08-24
799:2019-02-04
774:2019-09-26
749:1905.02244
727:1908.01748
706:1707.07012
685:2018-05-22
658:2019-11-25
631:2018-05-22
606:2019-11-10
576:1602.07360
555:2018-05-22
527:2018-05-22
518:"Students"
502:2018-05-22
477:2018-04-07
452:2018-04-07
422:2018-05-22
356:2019-10-02
328:References
322:ultrasonic
249:processors
211:SqueezeNet
197:Technology
111:Key people
80:2019-10-01
819:The Verge
596:The Drive
296:Autopilot
129:deepscale
848:Archived
769:EE Times
600:Archived
550:EE Times
472:EE Times
447:EE Times
397:EE Times
653:Fortune
417:Fortune
314:cameras
291:Fortune
245:sensors
238:Product
183:Visteon
161:History
124:Website
78: (
73:Defunct
52:Founder
42: (
34:Founded
253:NVIDIA
744:arXiv
722:arXiv
701:arXiv
571:arXiv
318:radar
310:LiDAR
286:Tesla
191:Tesla
105:, U.S
844:CNET
351:CNBC
320:and
306:Musk
301:CNET
281:CNBC
256:GPUs
247:and
185:and
169:and
88:Fate
44:2015
37:2015
260:ARM
177:in
131:.ai
117:CEO
870::
846:.
842:.
817:.
792:.
767:.
678:.
667:^
640:^
624:.
598:.
594:.
548:.
536:^
520:.
495:.
470:.
445:.
431:^
415:.
394:.
374:.
349:.
335:^
316:,
857:.
827:.
802:.
777:.
752:.
746::
730:.
724::
709:.
703::
688:.
661:.
634:.
609:.
579:.
573::
558:.
530:.
505:.
480:.
455:.
425:.
380:.
359:.
119:)
82:)
46:)
Text is available under the Creative Commons Attribution-ShareAlike License. Additional terms may apply.