You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Copy file name to clipboardExpand all lines: mian-shi-gong-si/ebay.md
+22-32Lines changed: 22 additions & 32 deletions
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -35,17 +35,7 @@
35
35
36
36
1. Can you describe your most challenging project and what steps you individually made to succeed in this project?
37
37
38
-
Customer engage our one product (highsigma analysis) and find our current solution is slower and accuracy is not good. After communicating with customer, find our current algri 
39
-
40
-
 Redesign highSimga application. Decouple application from FE and BE part to improve multiple process scaling up to 200 cores. To resolve high dimension accuracy and performance, combine importance sampling and machine learning method to predict circuit yield. Improve 3-5x speedup under high dimension situation.
41
-
42
-
 
43
-
44
-
 
45
-
46
-
 
47
-
48
-
 
38
+
Customer engage our one product (highsigma analysis) and find our current solution is slower and accuracy is not good. After communicating with customer and gather more information about user cases, find our current algorithm can't work well for high dimension situation. At the same time, we found application has strong couple with FE and BE, parallel scaling and job schedule balance is not good. Because evaluation schedule is very tight (three month). We made detailed improvement plan include prototype verification considering machine learning model to filter high dimension, redesign whole flow to decouple application from FE and BE to improve scaling. I setup weekly target and milestone. I lead two engineers to work on this project and setup daily sync-up meeting. I mainly focus on machine model evaluation and high level architect design. Finally, we combine importance sampling and machine learning method (lasso model) to predict circuit yield. It can resolve customer high dimension problem and get 3-5x speedup for all engaged cases. At the same time, new design can get linear scaling speedup up to 200 cores. We win the customer business. 
49
39
50
40
1. From a scale of 0 to 5, 5 being the highest – what number will you rate yourself in the following areas of experience to overall professional industry peers **(it is not a requirement to have experience in all category):**
51
41
@@ -55,47 +45,47 @@ Customer engage our one product (highsigma analysis) and find our current soluti
0 commit comments