Data, Structure, and Science in Hiring at Google | Kerry Cathcart, Staffing Program Manager

 Here is the link. 

Invest early - hiring is the art

data analysis, scientific approach for hiring 

New research and new experience

Four E: 

1. process is efficiency, effective, hiring right people
2. candidate - user experience
3. process - equitable - fair, not biased

Hiring committee

Four attributes:

1. Role related knowledge - day one - varies, level, communication skills, code - day one 
2. General cognitive ability - complex problem at workplace, figure out root cause 
3. Leadership - being a great leader, ability to lead a team - take ownership at work, influence the change, rally the peers around, see a problem and speak up 
4. Good or bad on behaviors 
5. Google culture unique and good ...

6:59 PM/ 40:19 

ownership 

Googleyness 

- Google culture - concrete - what to look for in the interview process 

culture fit - not really explain why

1. behaviors - 
2. seek out feedback to improve
3. ambiguity - wait for more info, or figure out what to do

13.05

Two kinds of questions to ask 

Behavioral 

Hypothetical 

Behavioral 

- Past focused

- Often begin with "Tell me about a time when "

- Useful to understand the impact a candidate has had in previous roles

Hypothetical 

Future focused

Often begin with "Image that..."

Useful to assess capacity for innovation in Google context

Dig into ...

What steps 

What resources

What do you do differently? 

Feedback - team member does not perform 

Drill up the information

What kind of information you gather? 

Think about assess the candidate

Rubics - assess 

Prioritize a portfolio of projects, cross organization 

Trap - first decent candidate to see

General hire or no hire - 

very detail interview feedback - take notes of specific answers 

Test candidate to get things done - company wide initiative 

20:00/ 40:19

Making hiring decisions by committee 

Benefits for Google long term 

Double down this - unconscious bias training - fairly and consistently 

My team - senior lead - feedback and continue to improve 

Keep the hiring trend and leadership - hiring is every one's job, not recruiter's 

Senior leaders - hiring committee - quality of work

Research shows three things to improve: 

1. Duration of the process, how quickly it can give the offer to the candidates
2. Candidate knows what to expect and how to prepare for the interview
3. Candidate knows how to make impact for the company 

24% -> 48%, talk about experience about family and friends. 

Leverage you can improve - 

what to expect and how to perform in the interview 

Really benefits - concrete how to make contributions 

This is really you can do for us. 

How to measure candidate's experience? 

80% - not offer a job - recommend to the friends - 

Network Flows: Max-Flow Min-Cut Theorem (& Ford-Fulkerson Algorithm)

 Here is the link. 

I like to learn those algorithms quickly again. It is tough job for me to learn how to talk and think in the graph algorithm - Network flows, Max-flow min-cut theorem. 

Flow Networks: https://en.wikipedia.org/wiki/Flow_ne... Ford–Fulkerson Algorithm: https://en.wikipedia.org/wiki/Ford%E2... Max-Flow Min-Cut Theorem: https://en.wikipedia.org/wiki/Max-flo... Proofs: Reference "Algorithm Design" by Jon Kleinberg and Éva Tardos Chapters 7.1, 7.2 for excellent proofs on all of this.

Things I'd Improve On This Explanation (w/ More Time): 1.) I should have done a walk-through showing how the residual graph dictates how the original graph's edge flows (f(e)) are updated each iteration. (That would've made it more clear how the residual graph in the Ford-Fulkerson algorithm tells us how to update the flow on each edge (f(e)) in the original graph along the s-t path P, THEN we update the residual graph (also along P) to prepare for the next iteration.)

2.) Go into the actual augmentation once we find an s-t path P in the residual graph. We can only modulate the flow f(e) for each edge in the original graph on path P ± the smallest value residual graph edge on path P. The smallest forward edge on path P in the residual graph is the "bottleneck" to how much we can increase flow along the path P in the original graph. (hard to visualize...the textbook may have to take it away with this one, but when you understand this you'll really get it after watching this video)

I also didn't talk about time complexity, but the amount of while loop iterations is bounded to the capacity coming out of start node 's'. We can't ever push more flow from 's' than the sum of capacities of those exiting edges. If each interaction increases the value of the flow v(f) by 1 (and v(f) starts at 0 in the beginning since no "water" is going through the "pipes"), we can do at most C augmentations of the flow network where C = sum(edge capacities leaving 's').

In each while loop: - O(|V| + |E|) to find the augmenting path - O(|E|) to update the flows in the original graph - O(|E|) to update the residual graph