Andreas Mueller’s  “Intro to Machine Learning” was packed and effectively sold out.  But his notebooks are on github and are the basis of an upcoming book on the Scikit-Learn.

Max Kuhn, Pfizer R&D,  and author of caret gave a “Workshop: Rule-Based Models for Regression and Classification”.  I really liked his insights on Quinlan’s C5.0 and Cubist. I need to get Khun’s book on Applied Predictive Modeling.

I kept hearing a lot of positive feedback on Chris Bergh’s of DataKitchen talk “Superpower your Analytics team with Technical Agile – A How To Guide” .  I’ll have to sit in on it next time.  Peter Bull of DrivenData also talked about “Data Science is Software: Developer #lifehacks for the Python data scientist” (GitHub).

Byron Galbraith of Talla, gave a great “Workshop: Bayesian Bandits”.  I captured some of the slides, and he has a GitHub repo on the topic.

Kevin Robinson gave a “A tour through the TensorFlow codebase“.

Kate Saenko gave a talk on “Adaptive Deep Learning for Vision and Language”, an academic summary of neural networks for automatic captioning of images and videos.

Janos Hajagos talk “Using Open Data with APIs” focused on open sources for working with health data. I love it when I can see the code. Beyond the Medicaid data sources and NY’s Health Data in the notebooks he pointed to Propublica‘s Drug Prescription Database and Socrata’s SODA.

Alex Felmeister, of The Children’s Hospital of Philadelphia spoke on “How Modern web technologies, data visualization and open data integration tools can transform research in childhood cancer”.  The talk (and slides) covered how The Children’s Brain Tumor Tissue Consortium, working with the  Department of Biomedical and Health Informatics (DBHi) was working with open source to solve rare deseases.  They have a GitHub repo of their infrastructure and models.

Richard Tibbetts and Alex Coventry of MIT and Empirical gave a talk on BayesDB (GitHub).  It looked very interesting.

Allen Downey, Olin College “Workshop: Bayesian Statistics Made Simple” was packed and they had to create two overflow rooms for the talk.  Again, loving the working code, and I need to get a copy of Think Bayes.

Alexandra Barker and Jeff Meisel of the US Census Bureau gave a “Workshop: U.S.Census Bureau API & CitySDK: Bridging the Data Gap”.  I’ve had to pull in public data sources for Kaggle competitions in the past, and this talk was a treasure trove of data and tools I could use, including API’s such as: Census Bureau API’s , a Python wrapper to the Census,  FactFinder , CitySDK (GitHub), mapping applications: OnTheMap ,  CartoDB , and example projects at the Opportunity Project.

I would have liked to attend Max Margenot and Jess Stauth‘s talk on “A Brief Primer on Data in Quantitative Finance”, but there is a Quantopian algorithm of an equity Long short algorithm they presented.

Charles Givre presented “Merlin: The ultimate data science computing environment”.  Trying to solve the sysadmin problem of the data science stack.  It reminds me of the Kaggle containers.

A lot to absorb!  A lot of code to play with!

I don’t know why I gravitate to long lists of maxims, as if the right combination could somehow unlock the path through a maze, but Ken Auletta‘s recent book “Googled: The End of the World As We Know It” (NYTimes review) omitted a chapter on lessons from Google and the changing media landscape:

1. Passion wins: “don’t settle”
2. Focus is required: the key to Google’s success is focused passion.
3. Vision is required: [have one!]
4. A team culture is vital: you want to empower the people that deeply understand what they are doing.
5. Treat engineers as kings: they are the creators.
6. Treat customers like a king: users come first.
7. Brand often means trust: as soon as someone extols ‘the brand’, they are out of ideas.
8. Every company strives to take the risks out of capital: “you need to control everything”
9. Every company is a frenemy: there are no allies, only interests.
10. The speed of change accelerates: the swiftness of change makes decision makers insecure, breeding fear and paranoia.
11. Adapt or die: smart companies focus on offense.
12. “Life is long, but time is short”: don’t be anchored down, and move very quickly.
13. A “free” web is not always free: the web needs another revenue stream.
14. Digital is different – when a product goes […] digital it changes fundamentally and irreversibly.
15. Don’t think of the web as another distribution platform – it’s many of them
16. Technology provides potent new targeting tools – from predicting behavior, action, and intent.
17. The web forges communities and threatens privacy
18. Beware the Government bear: when the Government plays referee, its decisions matter.
19. Paradox: the web forges both niche and large communities
20. More media concentration, yet more choice: giant media companies get bigger, but weaker.
21. Luck matters:  [so position yourself for lucky accidents]
22. No more old media magic: so media will die, some will be transformed.
23. No more new media magic either: some new ones will die too!
24. Don’t ignore the human factor: “truth is a liquid, not a solid”
25. There are no certitudes: and no company is invulnerable.

Of course the stories behind these maxims are interesting, and some are new to me.  If they catch your eye read the chapter because I can’t do it justice here.  The maxims that really matter are the ones I learn for myself, but it’s interesting to peek at someone else’s.

More than a decade later, the NetHeads vs BellHeads dichotomy has resurfaced and been recast. This partisan debate simmered after the .com and telecom crash; and the polemic seemed no longer useful as the underlying technologies evolved (to IP). Some even called for banishing this obfuscating metaphor. There was a brief reemergence when CellHeads vs. AirHeads fought it out for metro broadband wireless dominance. This too subsided as Wi-Fi hit the bottom of the hype curve; and WiMax sits there now. On the policy front people debated net neutrality.

But Professor Kevin Werbach clarifies the dichotomy as one really of business models and resulting approaches.

The Monists, including most network operators, see the infrastructure as the linchpin of the communications ecosystem[…]. The Dualists, […], see the infrastructure as simply a means to reach the applications, content, and communications on top of the network, which are the source of real value.

Bottom line?

All that is clear is that, sooner or later, more decentralized and collaborative models will sweep telecommunications as they are sweeping so many other industries and disciplines

And where do people stand?

Academics are split between the two camps. Anecdotally, the majority of economists seem to adhere to the Monist view, while the majority of legal scholars appear to favor the dualist perspective. Engineers are divided, often based on whether their roots are in studying telecommunications or data networking.

It will be interesting to see how this plays out. Let’s check in on this in 2016?

1. set realistic goals
2. encourage their employees to work as a team
3. and if they fall short occasionally, mine the failure for valuable lessons

No more, no less. Having recently scoured the business literature for useful lessons in leadership and organizational structures, I appreciated highlighting what should have been obvious.

Successful strategy of value creation in the startup world is simple: it uses a few clear guidelines and allows for flexibility, it is supported by organizational design, and it is not handed down from on high.  Kathleen Eisenhardt, of the Stanford Technology Ventures Program, has been speaking and writing about Strategy as Simple Rules for quite some time (HBR Jan’01).  A more recent talk (October 4, 2006) titled Research Lens on Understanding Entrepreneurial Firms really captures why you want to be in a low complexity and high velocity market with a great team.  The bottom line, predictably, is:

• To get started – you need a great team and a great market
• Raising money – is a negotiation
• You need to shape markets – power tactics
• Strategy as Simple rules – is an exercise in managing time and structure
• A top management team is the heart of great ventures!

But the the full podcast or short video clip explains why simple strategy doesn’t mean you should be winging it.  If you don’t have the time, cut to the presentation and my notes below. 

Quantitative research on entrepreneurship can be a useful source of success predictors for new ventures and Eisenhardt captures some interesting data.

How do you get started?  What does a really good team look like?

1. 3-5 people: 2 people have way too much to do but 6 or more bogs down decision making.  (the worst teams are 2 people)
2. Genuinely cross functional
3. People that have worked together: prior experience is really important
4. Teams with a variety of ages: they have different life experiences and can see different aspects of a problem

Bottom line, create a team with enough bandwidth to get the job done.

What Markets?

• Emerging: small, disorganized.
• Growth: $30M in sales, 20% growth
• Mature: > $100M-$200M in sales, growing slower

You want to start in growth markets.  Emerging markets are too early, hard to get the timing right, biz, tech, and customers are not quite right, and those mistakes will way you down.  It’s very hard to grow in a mature market.

The surfer model of venture growth:  three is a huge multiplier for a great team in a great market.  Great teams don’t find great markets, people start what they know.  Where you start is largely a matter of luck.  You don’t have control over the size of the wave.  It’s hard to recover from a poor start, while competitors compound advantages.  It takes a great wave and a great surfer to get a 20x business. 

What money?  Raising money is a negotiation.  When do you raise money? What does successful fund raising look like?  You want to raise it fast, enough, and get the most helpful investors. When?

1. When you have substantial signals that you are a good firm.  You have leverage when you have accomplished something. 
2. It is important to be realistic as to who you are and what you really have.
3. Find who’s funding ventures like yours
4. Go with regional investors.  (The power of casual dating:  find reasons to talk to people in the investment community before you need the money: hiring choices, product positioning, etc.)
5. The toughest challenge is closing the deal by having a credible alternative.  It’s about getting the “yes”.  There are different strategies for different people.  You trade on your status.

How do you create a market?  How do you find opportunities?  Great opportunities are shaped, not discovered.  You don’t find them, you socially create them.  You try to be come the cognitive referent in a market (the Vonage of VoIP, the iPod for MP3’s, etc.)  You want to give people a referent to understand your product.

You need a compelling story about your company; a memorable personable creation myth.  Make the media your friend; the media focuses on overcoming adversity.  Let the media work for you.

Protecting your market: You want to think about who might enter your space and how you keep them out.  Give them a reason not to by doing a marketing alliance or letting them invest in you.  In the negotiation you can delay the entry of others.  Finally, control the market and buy out others.

The special case of network businesses:

1. You need a network of complementors.  Build this network early and co-create the product with key partners.  For example, in the wireless gaming market you need relationships with carriers, handset makers, brands or content, and platforms (Sun Java, Qualcom BREW, etc.)
2. Build the whole network simultaneously to be the broker of those relationships. Not serially.
3. When there are uncertainties in the market, play all of them to find the answer fast.  Figure out the market with a lot of bets (e.g. genres of cell phone games).
4. You never tie up with competitors; too much conflict of interest.

Competing on the edge of chaos.  There is something about the amount of structure in an organization that is important.  Strategy is about having processes for product development, and the amount of structure is important.  This strategy is the collection of simple rules of managing time and structure. Sequencing these processes up the chain is important.  You ought to be thinking about these rules because there is an optimal structure depending on what market you are in.

Dynamic markets, or markets where high tech companies tend to be in, have 4 attributes: they are fast moving, complicated, ambiguous, or uncertain.  You don’t want to be in complicated markets (e.g. biotech).  You want low complexity and high velocity markets.  If you are in an uncertain market you typically want less structure and you want to manage the amount that you have.  This is one of the reasons why big companies have trouble working in new markets. 

You need to balance structure and ambiguity: most of the best new companies have more structure than their peers.  Most of the time startups are winging it more than they should.

• BBC Flying Penguins
• Google Gmail Custom Time
• Google GDay future search
• Google Book Search Scratch and Sniff
• Virgle: Virgin / Google JV to put a colony on Mars
• CrunchGear Verizon Buys ATT
• CrunchGear HandSolo (I guess this future technology finally launched at CITA!)
• TheUnFunded.com, where VC’s diss startups
• i’KIA, Kia unveils flat-pack car concept
• Free Nokia 6630 in magazine insert
• Defendius Door Chain Lock
• BMW Canine Repellent Alloy Protection (and past years)
• The Sun reports that President Nicolas Sarkozy will be stretched five inches to help him see eye to eye with supermodel wife Carla Bruni.
• Tribune Company changed its name to ZellCoMediaEnterprises and will print in licorice.
• Nestle Gives Candy Lovers ‘The Finger’
• Virgin Blue’s no chair faire
• NASA’s Space Station Robot Asks to be Called “Dextre the Magnificent“
• The old Water on Mars trick from NASA is back

And there are many other roundups of fools’ posts. None of these seem to make it to the top 100, but did I miss your favorite?

Peter was in town to interview Guy Steele, Dan Weinreb and others.  I liked how he let the community select and sort the list of top coders.  This is a truly interesting list of people; but Peter also acknowledged that the people who ended up on the list had to attain a certain amount of popularity of visibility to make the list.  I think that there is something to being a great hacker.

In the mean time, in a move to open source her book, I just noticed that SLammers has now posted her previous interviews on a programmers at work blog.

More interestingly, it highlights how Gavin Potter rose to the top ranks by linking movie predictions to human behavior by using ideas by Tversky and Kahneman to account for biases in how people rate movies [via].  It’s great to see that focusing on only the algorithm is the wrong way to go, and that the user interaction model is just as important.  Because the goal here is not just to get the prize for the best algorithm, but to get the chance to capitalize on sorting out demand.

What has brought on all this innovation is the structure of competition itself.  Netflix last year awarded the progress prize of $50K to the BellKor Solution that combined 107 complimentary models.   In writing how they achieved this progress, they explain how the spirit of collaboration and openness played a key role in the enthusiasm for the competition.  WIRED suggests this started when Simon Funk first described his method to attain third rank for a time.  This was further encouraged when the top ranked compared approaches at KDD 2007 conference.

This seems to have brought more focus on innovation through competition.  There has always been the KDD Cup, the first and oldest data mining competition; and there is the 2nd International ACM Conference on Recommender Systems in Lousanne, Switzerland on October 23rd, 2008.  But in an odd twist, some competitions are about ferreting out the best business plans: Strands introduced the $100,000 competition for the best early-stage project in the area of recommendation technologies, considering the technology, business opportunity and team behind the project; Textwise introduced the $1M SemanticHacker challenge for to anyone who can come up with the most profitable semantic web application, effectively outsourcing business development.

The number of entrants and finalists were as follows:

Judges were asked to evaluate all pitches against four (4) criteria:

1. Market Identification: opportunity, size and attractiveness, target customer identification.
2. Solution: value proposition, differentiation, financial viability / profitability.
3. Presentation: clarity, articulation, passion.
4. The big question: how likely are you to commit resources to this proposal? Would you fund them?

We could use the following scores: 0 (spinning wheels), 4 (has traction), 6 (going places), or 10 (rock solid). This helped separate the top entrants from the middle ground. The judges’ results varied substantially on any one criteria, but after a discussion we agreed on the top entrants.

The key to the pitch is not to provide the complete story, but simply to convince the person to have a follow-up meeting; and for this you need the hook that has them coming back for more. Sean Wise adds that you need to: 1) lay out the pain statement, and 2) you must show the value proposition, or how your venture solves a problem. A successful pitch must be succinct, easy to understand, greed inducing, and it has to be irrefutable. And yes, if it is a real pitch, you need to close with a call to action for a next step.

There are lots and lots of resources on the web, including an elevator tv channel and the Dragon’s Den, but as someone on the receiving end of 65 pitches over a couple of hours these are some more suggestions.

First, state the problem. This contest took place in front of a large audience in a theater like classroom. Too many times the presenter would open with: “How many of you have had problem X?”. This kind of opener would never be used in a real elevator pitch and is a waste of time because the presenter then has to wait for an answer from a non responsive audience.

Then, state the market. Be specific, show you know your stuff and that you have thought deeply about the problem and the customer. This event allows for questions, and if you get a question it’s a good indicator that you have a judge engaged.

Introduce yourself and your innovation. Provide your name and company name. Wear a badge. Come to the event with business cards. Have a web site up before the event. Don’t introduce yourself up front. I find I will not remember a name unless I have something to attach it to; in this case the pitch. Then articulate the solution you plan to bring to market.

Finally, shoot for the big question. State what you are asking for! Say what you need. A number of pitches identified a problem and a solution, but never stated what they wanted!

I found humor a tricky tool. Humor can really work to make a pitch memorable, and this is really important. But it also takes time and creates noise in the audience; which interferes with getting your message through to the judges. So don’t overuse humor, and use it at the end of the pitch.

Also, don’t memorize the pitch word for word. You need to know the storyline and the transitions points, but not every last word. One guy had a very good pitch; it was very tight. But at the final round he choked and could not get back on track to finish the pitch!

What is interesting to me is the history of how BladeLogic was started.  The creation myth is that back in 2001 sysadmin Vijay Manwani had an aha moment when trying to solve the problem of configuring many datacenter boxes.  The reality was that Vijay, already a CTO at a number of previous companies, was an entrepreneur in residence at Battery Ventures and identified a company called Network Shell, which had an innovative product that customers were already buying, but lacked the management team with the industry expertise to scale the business.  What is interesting is how a small tool like nsh grew to the industry sector category of “datacenter automation” (DCA).   This seems like a classic case of lead user innovation.