jump to navigation

How to Humanize Data for Enhancement (Notes from May 20 – 26, 2019) June 12, 2019

Posted by Anthony in Automation, Blockchain, Digital, experience, finance, Founders, global, Hiring, Leadership, NFL, questions, social, Strategy, Uncategorized, WomenInWork.
Tags: , , , , , , , , , , , , , , , , , , , , ,
trackback

I’m going to keep this brief this week, as I’m trying out an added format of posting. I believe I’m going to test Nuzzel news to connect and share the links of articles I’ve read through the week (maybe once or twice – weekend one out on Monday, and another for the week through Thursday?). This post, concentrating on my week’s podcast or segment listens, will still go out. That’s my goal, though.

The week I listened to these, I was actually on my way to Monterrey with my family, though, so I had a solid 90 minutes each way to go through some longer episodes – thankfully, I found Doug’s take on culture and leading a community with Eat Club very insightful, as well as a blockchain discussion by the a16z squad particularly fun.

With the news and media focused on AI, Bias, Diversity and Ethics of ML and scariness of algorithms, it’s good to hear opinions that are focused, but thoughtful on the premise of what is being proposed. In a vacuum, yes, these can be dangerous. However, that’s not often the case when models are being deployed into production. Some are unintentional or started with a simple question of relationships (think Cambridge Analytica and how it started via Likes research on music at Facebook). That’s not necessarily justification for something that clearly had a large impact, but we’d like to know that teams are improving and have that awareness now more than ever before. That’s a positive.

I believe that these episodes, though, take a bit more of a nuanced approach of mathematics and numbers and attributes a human eye that has acquired the necessary expertise to design better models or come up with a better framework for action. We hear that with Dan Waterhouse of Balderton Capital (of a delightful Applied Math degree), who had to learn to apply more of an operational and human/market behavior investor since he had the metrics focus while learning. Then, a fascinating segment with Laura Edell, a sports fan, who struggled to believe that the metrics that were widely accepted as ‘most accurate’ (re: not necessarily max accuracy) included all components of what could be measurable – made some assumptions to test based on twitter/media analysis and coupled it with the data that was widely acceptable to come up with an excellent Final 4 Model.

To continue the theme, not to be outdone, Geoffrey Batt and Nikos Moraitakis each spoke of differing aspects of metrics that they honed in on and ignored the common threads of why those weren’t valid, despite what they saw as opportunities. Geoffrey as it pertained to Iraq investing and Nikos in the form of starting and building a company in Greece, not exactly the traditional hotbed. However, each persisted and are successful today – so there’s a lesson in doing the things that we want to do.

Remember, ultimately, numbers and averages usually take the aggregate, collection of those that may not have had the timing, willingness or ability to push but with enough persistence and support, some will certainly persevere.

Hope you enjoy!

  • Doug Leeds (@leedsdoug), CEO at Eat Club (Dot Complicated, Wharton XM)
    cropped-logo-highres1

    • Talking about going from Ask.com to Eat Club – other companies, as well
    • Quiz leading in was on different likes/dislikes
      • Harry Potter-themed eating pop-ups
      • Anti-spoiler pin (digital, season and episode to avoid conflict)
        • Anything was good that brings community together
        • Said that there were employees at work that took off Monday after GoT finale since they hadn’t watched
    • Delivery people are employees, not contractors – become regulars at corporate events and a part of the culture
      • Help build the culture of the locations – food and how they match
      • Building culture of Eat Club with improv – ability to improve skills there, as well
    • Eat Club numbers – 1000+ companies served, about 25k lunches provided
      • Regional difference of Indian focus in NorCal, Mexican focus in SoCal
        • Acquired an Indian food provider for new techniques – Intero?
        • Acquired another, as well
  • Five Open Problems Toward Building Blockchain Computer (a16z 5/16/19)
    ah-logo-sm

    • Ali Yahya (@ali01), Frank Chen and looking at crypto
    • New paradigm has to improve upon one or two particular ways for new applications, but likely sucks in most others
      • Mobile phone as enabling behaviors for instagram/uber and such with camera and gps in the phones
      • Have to trust the company currently to do what they claim to be doing – trust Google with so much to have the interactions
      • Now, have a computational fabric that separates the control of human power, self-policing and security bottom-up
    • Difference in communication cost – bounded on the end by speed of light
    • Trying to make networking efficient – miners propogating to other miners – blockchain distribution network
      • 1 MB vs 2 MB blocks but can kill some small miners
      • Agreeing on blocks or updates are final
    • Non-probabilistic vs fast – need to be faster than 60 minutes, for instance
      • Improvements on networking layer with Ethereum 2.0, Cosmo and cost or Proof of Stake (vs Of Work)
        • Who gets to participate? PoW requires everyone to compute an expensive proof of work.
        • PoS – intrinsic token that you have to own in order to buy participation, 2% ownership says 2% of the say to say yay or nay
        • Cost of participating is less expensive because it’s intrinsic
      • Practical for everyday use, small interactions between people and machines or people
      • Trust as the bottleneck to scale
    • 3 Pillars of Computation/Scalability: Throughput, latency, cost per instruction
  • Daniel Waterhouse (@wanderingvc), GP at Balderton Capital (20min VC 091)
    cqis46nlxvy5bkvj15kh

    • Sits on boards of Top10, ROLI, Lovecrafts, TrademarkNow etc
    • Was 5 year partner at Wellington Partners and invested in EyeEm, Hailo, Yplan, Bookatable, SumAll, Readmill (sold to DropBox), Qype (sold to Yelp)
    • Break in 1999 by getting to work at Yahoo for about 10 years before getting to venture in 2009
    • Applied Math background, tend to understand complexity and appreciate solutions in technology – different experience for entrepreneurs
      • Learned to be less metric-driven – less obsessed on numbers and data, more human and market dynamics
    • Operators vs career VCs – successful firms take different skills at different times
      • Exit, raising money, portfolio management, scaling, growing
      • Important to watch how a business is grown, whether it’s yourself or by being a long-term investor or close in other ways
    • Entrepreneurship as a career path and as a global one, investors in similar mindset and ambition – can take it anywhere
    • More seed capital and smaller checks than later in Europe at this time – maybe more competition higher stages
    • Thinks that consumerizing SMB software has a lot of room to grow, as well as enterprise software – easier UI-driven tools
      • Developments in AI – mentioned real-time trademark analysis (TradeMarkNow, for instance)
      • How to give great recommendations after gathering consumer tastes and insights
    • Book: The Brain That Teaches Itself – neuroplasticity of the brain
    • The first $10mln you invest, you’ll lose – advice
    • Overhyped sector: food delivery; Neglected: apply behavioral science to tech problems
    • Favorite blog/sector: longs for better content – thought TC was great when it was started, but now it’s Medium individual articles
    • Most recent investment Curious.ai – made it in 40 minutes & ambition was compelling
  • Geoffrey Batt (@geoffreysbatt), Nature of Transformational Returns (Invest Like the Best, 4/9/19)
    • Iraqi equities – history of asset classes
      • Time and patience is a big factor for how you look for results
      • Japan, for instance, 1950 – 1989 100,000% return but now is still below the peak from 1980
    • May have a decade of not working out return-wise
      • Nothing to show for investments, incredibly challenging, especially now – hedge funds or LPs
        • Reporting weekly or daily basis, long-term investment could be 6 months or 1 year – pulling money after first issue
      • Hard to ride out periods of long-term returns (think: re-rating countries)
    • LPs as agreeing to the time horizon – maybe investment committees that are making decisions on career-risk for institutions
      • Iraq to one of those – not likely to get paid off if it works, but if it doesn’t, get demoted or fired
      • Had to approach HNW and family funds – adjacency risk for people next to seed investors (if one is weighted on fund)
    • As student, he majored in psychology – said there was a guy in the seminar class that was a unique thinker
      • Daniel Cloud had co-founded a fund that invested in post-Soviet Russia before doing his PhD
      • Asked Geoffrey to come work for him and learn investment markets
    • Wanted to find next big thing – first, find a place that everyone thinks is awful (in 2007, this was Iraq)
      • Does perception meet reality? How many people dying in the war every month?
      • Country portrayed as “failed state” but oil production was increasing, CPI was 75% yoy but now 5-10%
        • Currency appreciating against the dollar, now civilian casualties in a month were down to 200 from thousands
      • Normal visit to Baghdad – mundane here, out-of-place was that 2 guys in middle of afternoon were playing tennis, kicking soccer
        • Visit companies, stock exchange, meet executives, go to a restaurant, how easy is it to hail a cab, get around
        • Critical process of infrastructure – are people paying in dollars (bad sign), local currency?
    • News media is just not set up to convey complexity to the audience – alienate, progressivism as a service, readers change
      • Experts don’t have skin in the game so they don’t face career risk – betrayal if you suggest otherwise, likely (even if true)
    • On his first trip, mentions Berkshire – early 1990s where they first invest in Wells Fargo
      • Managerial skill in banking is paramount if levered 20:1 (make 5% mistake and you’re done)
        • Don’t want average banks at great prices – want great banks at slightly unreasonable prices (thought about this in Iraq)
      • CEO of first Iraqi bank – unsecured loans to taxi drivers – less likely to take out the loan “between you and I, I’m a cowboy”
      • Entrepreneurs in these areas have to be better than others because of the instability vs the stable environments
    • Stock and capital guys – stock may trade at 3x earnings or 4-5x FCF, top and bottom lines growing at 20% per year
      • Usually just put there as knowing someone powerful but they can be bigger as allocators
      • 2008-09: Size at the time of the stock market – traded 3 days a week, 10am – 12pm on a white board
        • 60-70 companies, maybe 20 suspended from trading because 0 annual report – $1.8bn, smaller than Palestinian market
      • High growth company that is super opaque – can’t meet or won’t meet with CEO, maybe some others that state-owned enterprises that just want to keep paying salaries; maybe 5-10 companies that are investable
        • Equally weighted these companies initially and was still learning – now, he developed relationships and is on the board for companies
      • Does he want to put more $ and concentrate on the companies that he trusts and follows the guide
    • Largest holding for them – Baghdad Soft Drinks (Pepsi bottling and distribution for Iraq) – mixed-owned enterprise in the 90s
      • Local businessman (Pepsi is the dominating market share – ~70-80% vs elsewhere where it’s split) saw it was mismanaged
      • By 2008-09, were going to default on the loan they were floated – businessman bought the loan, fired management & 2000 ee’s, switched equity
        • Within a year, it was fully certified, tripled production, profitable business
    • Now, 5 days a week but still 2 days a week – had foreign investor interest, $5-6bn, couple IPOs successful
      • One telecom has about $1bn market cap, 14.5% dividend yield, 33%+ FCF yield
      • Foreign money came in 2013 but ISIS scared them off – coming back and interest from banks – arranging trips and momentum
      • Key problem – no 3rd party custodians (compared to Jamaica or even in Africa, HSBC or anything) – working on getting one
        • Makes it difficult to bring in foreign money – exchange is the custodian (which is actually safe)
        • No margins, cash-based market and settlements of t+0, no short selling (can’t sell unless have stock)
      • Oil collapse depressed prices and ISIS issue but has been up over last 2-3+ years, cheaper today than when he invested 11 years ago
    • Multiple expansion is the question for returns – 1x to 25x or 4x to 15x – depends on what they are as compounding
    • Kobe Bryant complimented him after a junior year game in high school – already looked at as a superstar – saw Geoffrey was dejected
  • Nikos Moraitakis (@moraitakis), Founder, CEO Workable (20min VC FF024)
    yawnofc__400x400

    • VP of BD at Upstream, enterprise sales in 40 countries in 4 continents
    • Wanted to start a company with type of what they wanted – centered around product, engineering company, starting in Greece
    • SMB over Fortune 500 because of product-focused and not corporation or enterprise-based
    • Applicant-tracking systems (ATS) – keep track of arcane process, don’t want to touch things – collaborative processes
      • Simple interface to solve problem of recruiting model (which is still 50+ years old)
    • Problems aren’t location-based – they’re conceptual – designing product, PMF
      • Opening with $500k or 50-200k if needing to get started, not necessarily $5-10mln like other markets or tech hubs, industry
      • Hiring engineers that are good in Greece isn’t so much a problem if you have a good company and network to attract talent
      • Moved to US not for VCs (doing good business, VCs pay attention) but they had 50% of customers without having anyone in US
        • Wanted to start customer support infrastructure, services and otherwise
        • Talked about marketing or accounting businesses on tech that taught companies that it may be worth it to update
    • Metrics that he pays attention to: month-to-month above $2mln annual revenue, ratio of new biz and lost biz
      • Not celebrating fundraising – few drinks but says it’s like having new shoes at the start of a marathon
      • Talking about investors and the relationships built to work together
  • Jaz Banga (@jsbanga), cofounder, CEO of Airspace Systems (Wharton XM)
    airspace-metal-crest2x

    • Meeting Earl, cofounder and maker, at Burning Man – drone being annoying above him while he did tai-chi
    • Prototyping the interceptor drones after seeing military request proposals – had an issue with drone over nuclear subs
    • Can place thermal scans and other security or rescue methods
  • The Transformer (Data Skeptic 5/3/19)
    • Encoder/decoder architecture of vector embeddings word2vec into a more contextual use case
    • Keyword lookup may not work (using ex of bank – river bank vs financial)
      • Humans at 95%+ accuracy, computer maybe around 50%
      • Encoding as correct interpretations and weights for context – emulating process
  • Laura Edell (@laura_e_edell) from MS Build 2019, NCAA predictions on Spark (Data Skeptic 5/9/19)
    build-2019-intro

    • Been at Microsoft for 3 years, supporting Azure for all their customers – quantum physics and statistics background from school
    • ML in cloud for customers – don’t know why they want it, just think it will be useful
      • Ex: image recognition (on techs with bodycams), HVAC documentation and augmentation or signal processing in anomalies of wave patterns
        • holoLENS use case
      • Played 2 sounds for Build presentation – her son blowing on her arm vs HVAC system custom sound – training sets and transfer learning
    • Can take a few real image and do a bunch for training: rotating, zoom, Gaussian blur, cut out background
      • Sound – same: take out environment, pauses and silence
      • Can turn 10 images / sounds into 30-50 per class
    • Active learning: model over time that can train itself and then retrain itself
    • Business domain expertise – her Final 4 model
      • #1 feature was wins, 2nd SoS, 3rd home court advantage / location – let machine validate the expertise
      • Validation of revenue drivers from machine – more importantly, if the opposite occurs – revenue doesn’t agree with data features
        • Used statistics to train data from the start in football, sports data – ncaa – teams, tourneys, prior history
          • Brought in her assumption of player-social index where they scraped sentiment and video analysis for team effects
        • Chose to use Azure Databricks (Spark background), store it in Blob and only retraining on Just-in-time in a Docker image
      • ML Flow, set of score for model – training set and .py and score.py or source data gets grouped together in Azure
        • Docker pulls them together easily and image is built, Azure DevOps can do VC
Advertisements

Comments»

No comments yet — be the first.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

%d bloggers like this: