Raising a Neural Net

I’ve been a little busy over the past year or so. You see, I have been creating the most advanced neural net known to mankind. It took me 9 months to code it and get it just right. My husband helped to provide some of the code, I provided the rest. As often happens, we repurposed some of the code from people we trust. It’s basically open source. Almost anyone can do it. Compiling it wasn’t the most fun I’ve ever had… There were some late nights, and I definitely lost some sleep. But my neural net turned out really cute! I lucked out. I didn’t plan for it, but it has dimples!

Now that she is here, I’m training my model. It will take a long time to train. 18 years by some accounts. Maybe more? I often wonder if the training time depends more on me training it well or its implicit structure?  It has many layers and a variety of activation functions. There is definitely drop out automatically encoded. By all accounts, it’s very sophisticated. The only  major downside is that I have to feed it all its labeled and unlabeled data manually.

When I put together my training data, I think a lot about ethics and values. I want my little neutral net to eventually make good decisions and be kind to other programs. It’s not so easy to build an unbiased model. Because I’m creating the training set, it’s up to me (and my family) to teach it well. Luckily, it’s not a one and done situation. If, in a few years, I learn that my neural net has learned to hit other neural nets, then I can ramp up the training data to try discourage that activity. But, I guess ultimately, it’s still a black box. I’ll never understand exactly why it made each decision!

Having a human neural net makes me think differently about how I might train my digital neural nets in the future… Meanwhile, my neural net woke up from her defragging and sleep processes, so I need to go. I get to go to stack more soft blocks so she can collect more unlabeled data about gravity!

Posted in Communicating Math, data science | Tagged , , , , , | 4 Comments

Are you masculine enough to be a trusted leader?

I’ve been a woman mathematician in Industry for almost 5 years. I have Malcolm Gladwell’s requisite 10,000 hours to be an expert.  Throughout my experience, I have struggled to gain the trust of my more technical colleagues. It’s not a new problem for me, but, upon joining Industry, a new wrinkle to this puzzle has presented itself: Do I want to be trusted as a technical individual contributor or as a leader? I like both options! I am capable of both options. And, as a result, I waffle back and forth between the options.

Every time I’m on one track, I think fondly about how the grass is greener on the other side. I think, “Gosh, I miss sitting quietly at my computer and writing code.”  Or I think, “Man, I wish this project was better managed so my good work was used more!” On the whole, I have chosen to lead people. Because without a good leader, the technical work doesn’t get used. And I came to Industry so people will use my work instead of having the work live primarily in a journal article somewhere.

But, do ‘leading’ and ‘doing’ have to be so separate? In some companies, the two development tracks are presented as something which can be done together. “You can lead people AND write code!” they say. But in my experience, any leader who does this effectively is working 80 hours week. Spending 40 hours on technical contributions and 40 hours on leading people. And, ultimately, when time is an issue, an individual must decide which is more important to them: leading people or doing technical work.

And, based on some new research, there are other reasons to believe that these two options are NOT options which can be taken together. M. Teresa Cardador & Brianna Caza interviewed more than 330 engineers over the last 4 years [HBR article]. Taken together their conversations show that technical folks view managerial roles as undesirable. Teresa has done previous research on the prestige hierarchy of this highly technical space. Our culture teaches us the hard skills we need to be technical capable are separate from the soft skills that make us good with other people. What’s more, we “also learn that these skills are gendered, with the [hard skills] viewed as more masculine, more revered and higher status; and the [soft skills] viewed as more feminine and lower status.” So, as I move into leadership, do I have enough technical skills to be seen as trustworthy on technical topics? Am I masculine enough to be trusted? Without that, my value to the company is seen as lower than the individual contributors because I’m using my “feminine” skills to get work done.

“It seems like these things, these skills, these traits that I’ve honed for a very long time…one might label as soft skills maybe…are not really the kinds of things that get rewarded as much on day to day. Or are being recognized.” – Cardador & Caza

This quote is from the article, but I could have easily said the same thing.  I know many people who agree with this statement. Being a leader is a burden and is ‘less valuable’ than being a technical data scientist. Devaluing leadership isn’t really a problem, until you layer in the gender bias. Cardador & Caza found in their study “that while some women pursued these technical supervisor or management roles based on their preferences, some were also mentored into these roles.” So, women are getting pushed into these roles despite their other preferences.

“When women disproportionately occupy roles that are less valued or unwanted, it can reinforce stereotypes about female engineers being less technically skilled, make them feel less respected, and create the illusion that they are not a ‘real engineer.’” – Cardador & Caza

And that’s exactly how the choice feels to me. Do I want to be a “real data scientist” or do I want to be a leader of data scientists? I find leading to be more personally fulfilling and, I believe, leading makes me more valuable to my company because I will insure the work of the non-leaders finds its best use case. I spent decades of my life politely fighting with men to make them see that my hard skills are just as advanced as the men’s skills are. I spent these years metaphorically saying, “I’m masculine! I’m one of the guys!” But now, with a choice to serve the greater good and use the skills that are really underrepresented in tech (social skills), I am undermining all that credibility I built.

The perception is tough to shake. Cardador & Caza talk about resilience of women in tech. Mostly it’s about staying true to oneself and ignoring the peer pressure. It’s being able to say: “everyone else will think I’m making the less prestigious choice. And that’s OK.” But honestly, I can’t decide if it is OK, because if I lead, then the individual contributors will perceive that I’m more feminine and approachable and therefore less technically capable. And if the team I lead doesn’t believe me to be capable, then I will have a harder time leading the team effectively. And I don’t know how to solve that puzzle… So, the question remains: Do I appear masculine enough to be trusted?

Posted in Business, data science | 2 Comments

Oreo Preferences

Everyone loves Oreos! They are delicious and sugary and chocolatey and have excellent filling!

For your pleasure, I have done extensive research on the preferences of several different kinds of Oreos. Well, maybe not extensive research. But a bunch of research anyways!

I did a small study to compare Oreos, Oreo Double Stuf, and Oreo Thins. Here are some basic stats about the cookies in question:

Name Calories/

Serving

Serving Size Calorie/

cookie

Servings per box Cookies Calories per box
Oreo 160 3 53.3 12 36 1920
Oreo Double 140 2 70.0 15 30 2100
Oreo Thin 140 4 35.0 10 40 1400

I had 21 participants. Only 4 of my participants regularly pulls their Oreos apart. But 41% of the participants dunk their Oreos in milk! So the milk tradition is still going strong. Of these 21, 10 preferred double stuf, 8 preferred classic and 3 preferred thins.

I wanted to understand the overall preference between these cookie types versus their caloric intake. I want to determine if this data would change anyone’s mind about which cookie they eat.

I am a die-hard Double Stuf eater. I don’t understand why anyone would eat a classic Oreo. There is far too much cookie for the amount of filling. However, there is the disappointing truth that my cookie choice costs 16 calories more per cookie. And I only get 30 cookies in my box of cookies. If I’m trying to count calories the question is: do I love a single double stuf cookie 30% more than I love a single regular Oreo?  For me, there is no question. I absolutely do. Lets get into the weeds: if I prefer double stuf and my enjoyment level is an 7, but my enjoyment level of a classic is a 4. Then I enjoy my double stuf 7/4 -1 = 75% more than I enjoy the classic. So, even if I’m watching my calories, I shouldn’t switch to classic.

My participants were given a single cookie of each type as asked to rate their preferences in order (1st, 2nd, 3rd) and then to rate their enjoyment numerically from 0 to 10.  So I can also consider their answers. Do those who prefer double stuf prefer it by more than 30%?  Of the 10 people who prefer double stuf, only 5 prefer it by more than 30% of the value they gave classic. Thus, the 5 for whom double stuf is not vastly superior, they should probably stick to classic.

What about thins? An Oreo thin is only 50% of the calories of a double stuf and 66% of the calories of a classic. Of the 8 people who chose classic as their preference, do they prefer it 50% more than a thin? Only one person rated the classic more than 50% higher than they rated the thin.  Thus, 7 out of 8 people who prefer classics actually gain more enjoyment/calorie from an Oreo thin.  Additionally, 6/10 people who prefer double stuffs actually gain more enjoyment/calorie from an Oreo thin.

What about me? I was surprisingly pleased with the taste experience of the Oreo thin.  I gave the thin the same value as the double stuf: 7/10.  Therefore, if I want a single cookie, it’s actually WAY better for me to get 7 points of enjoyment out of a thin than a double stuf. As a result of this study, I buy Oreo thins instead of double stuf. What will you do?

In conclusion, although thins were the least desirable (only 3/21 people preferred them most), if you only get 1 cookie and you want to optimize your enjoyment to calorie ratio, then most people should buy thins. 16 out of 21, or 76%, of my participants would get more enjoyment/calorie from a thin than from any other Oreo.

 

 

Posted in Communicating Math | 3 Comments