Overview

Courses

Udacity - Statistics (ST101)

  • started_on: 25 jun 2012
  • on: 01 Aug 2012 - 08 BRule
  • now_at: 4.1
  • goal: 2013

Udacity - How to build a startup (EP245)

  • started_on: 01 oct 2012
    • Phase 1 - Get ready to sell
      • MVP - minimum viable product
        • meets minimal customer needs
      • Identify customers and markets avail.
      • JerseySquare example
  • on: 02 oct 2012 - Lecture 1.5b
  • on: 04 oct 2012 - Lecture 2
  • on: 13 Oct 2012 - Lecture 2.21
    • Value Proposition - satisfying a customer need
      • solve a pain & create a gain => MVP
      • product market fit at the intersection of prod/market
      • keep in mind to "get out of the building"
    • Customer Segment or market
      • model the product'st "Persona"
      • create Pain Killers - problem vs need
      • depending on market web/physical/etc you need a model to get feedback on your solution
      • Value proposition - why hasn't it been solved yet?
      • ayasdi - tech insight / twitter - market insight
  • on: 14 Oct 2012 - Lecture 3.1
  • on: 23 Oct 2012 - Lecture 4.1
    • Customer segments
      • understand whom are the people who might buy your product
      • there might different segments available
      • create a "day in the life" of a customer for a persona
      • customer gains - list of what can they win by using your product
      • you have to validate some of your hypothesis
      • customer pains - opposite version of gains, need to understand it
      • know everything about your customers - create a customer persona
      • who's the customer - in context - payer might be a different person
      • B2B present less chances to get useful survey results compared to B2C
      • 2 sided market, e.g. google - searching users / paying advertisers
      • 4 types of startups: existing market, resegment market, new , clone
        • changes everything: initial costs, market size, sales model
      • chasm between early adopters and mainstream market depending on startup type
        • wider depending on market type
      • revenue stream for each value proposition
      • first customers for startups are just crazy people like you
      • 85-90% features are unwanted & un-needed by customers
      • most fail due to a lack of customers not because of the product (also: vision, founders fight 1/4, too expensive to fund)
      • founders spend 20% of their time w/ customers
      • [4] distribution channels:
        • virtual / physical 2x2 matrix: 2,2 twitter, 1,2 amazon 2,1 ERP 1,1 cars
        • most successful startups pick only one distribution channel, at least initially
  • on: 1 nov 2012 - Lecture 4.5
    • physical distribution channel (4.5)
    • solution complexity: system integrators - handle the most complex solutions > Direct Sales > VAR > Retail > Web
    • as we go up solution complexity, the value added increases; higher volume are at the low end
    • channel economics
      • direct sales: e.g. $100 = cost of raw materials ($33) + r&d, sales & gen. admin ($20) + discount ($10) + PROFIT ($37)
      • indirect sales (reseller): cost of r.m. ($33) + r&d, sales & gen. admin ($15) = 48$ + discount ($10) + reseller ($20) + PROFIT ($22)
      • selling costs are lower, but losing more on total profit
    • Customer Relationships
      • how do you get, keep and grow customers? - double barrelled funnel the same, slighty different for web and physical products
        • get - viral loop, in-funnel / keep / grow, out-funnel
      • update the consumer archetype, how do they know you and research about your products, how do they buy?
      • have an opinion based on a fact, try to continually refine the archetype
      • pay - PR to promote your product or company / goog or fb pads based on archetype / trade shows / e-mail / mail /
      • unpaid - blogs, guest articles, social media, speeches
      • physical channel: awareness, interest, consideration, purchase
        • modify the funnnel to fit your needs, use it as a template
      • 5.11 - 2 nov 2012
      • getting customers is more expensive than keeping customers
      • keeping customers using loyalty programs, newsletters
      • grow customers - end funnel
        • up sell - different product lines w/ different prices / cross sell - selling similar products / referrals (physical)
      • get mobile funnel: acquire - how to get them to visit your site / activate - pay / sign up - engage w/ the product
        • 2-step process; customer aquisition cost (CAC) - 10k users, .5 ad word -ppclick, 500 sign up, 50 paying => $100 C.A.C.
      • churn: how many customers leave
      • grow for web: lifetime value (LTV), how much will they spend from get/keep/grow, we must have LTV > CAC
      • ideally LTV should be (much) higher than CAC
  • on: 6 Nov 2012 - Lecture 6.1
    • Common mistakes
      • revenue stream IS NOT the price, and a lower price than competition is not necessary needed
      • it must be driven by the customers
      • pricing must be based on value and not cost, customer should feel that
    • Revenue stream: the strategy of the company to generate cash from each customer segment
      • each revenue stream may get different pricing tactics
    • Pricing: the tactics used to set the prices
      • fixed: value based on customer segment, based on volume, cost + mark-update
      • dynamic: negotiable, based on yield, auctions
      • depending on your customers, get out of the building :)
    • Fee: proportional to the usage of the service / flat fee to access to service / licensing - fee for use IP
  • on: 7 Nov 2012 - Lecture 6.10
    • Market types affects pricing
      • how will the competition react, can we undercut them?
      • what are their costs
    • Single/multi-sided market: e.g. Google - cares about users first, revenues second
      • for a physical channel, go for both at the same time
      • for multi-sided market, talk to the payers not only the users
    • "revenue first" companies - focusing on revenue more than others
    • "Hockey stick" - startups in a new market, flat revenue initially, then after an inflexion point, exponential growth
      • 1st year bump - due to early adopters
    • "draw the diagram" on how you'll be making money
    • key revenue model questions
      • what are the customers paying for and how is easy is for them to keep paying
      • don't put all features in the initial product (packaging)
      • estimate the market size: CAC, LTV, how many units will you selling
    • get feedback from customers 100 web / 10 non-web
  • on:11 Nov 2012 - Lecture 7.1 - Partners
    • Partners: what type of relationships you need to have with them
    • shared economics, what's in it for your partner / mutual success & failure
    • working together with others to help build your company
    • strategic alliance: use partner to build a "whole product", using 3rd parties to provide solution to customers
    • greatest strategic alliance: iPod (apple + record labels)
    • joint business development: intel + OEMs => intel inside, promotion and ads programs
    • in new market no need for "earlyvangelists", as most tend to fail
    • coopetition: to do something jointly for their industry (tradeshows, associations)
    • key suppliers: outsource part of company to 3rd party - backoffice, manufacture
    • virtual channels/traffic partners: deliver predictable levels of customers (web)
    • risks: e.g. Boeing building 787 - outsource all manufacturing, supply chain too scattered - decisions made by accountants and not engineers
    • risks: longest of their decision schedule is yours now as well; no clear ownership of the customers; products lacks vision; difficult to unwind or end
    • investments from large companies: "what's in it for us?" - buy us without (investing) money
      • who's the sponsor? and what's the motivation?
    • understand that you're not a PEER to large companies
    • partners can be a potential acquirer
    • try to understand if customers want you to have partnerships
  • on 12 Nov 2012 - Lecture 8 - Resources, Activities and Costs
    • most important assets: finance / money, IP, physical resources, human resources
    • 4 critical: physical, human, financial, intellectual
      • physical goods are capital intensive, how to scale
      • easy to start-up, especially in softare, some rough calculations on capital invested vs. possible gains
        • VC in both operating and financing phases, angels, friends and kickstarter in financing phase only
        • lease-lines/factoring only in operating mode
  • on 26 Nov 2012- Lecture 8
    • HR: teacher: learn about a subject, improve a skill - coach, mentor - giving and taking advice
    • always hire people better than you!
    • IP: trademark - you can protect branding (marks, logos), copyright the 'creative work', contract/NDA, patent the 'inventions'
    • Costs (alywas < Revenue) in-sync w/ investors from day 1
      • fixed costs: building, employees - every month, don't move
      • what are the most important cost to operate the business
      • startups search for the execution metrics, but you need to first understand the metrics that matter
      • burn rate - how much money are spent per month
  • Reading recommendations:
    • Startup Owner's Manual
    • Business Model Generation
    • The Innovator Dillema / The Innovator Solution
  • goal: 2013

Udacity - Software Testing (CS258)

  • started_on: 27 jun 2012
  • on: 27 Jun 2012 - 1.11
    • python -O : optimize, disable assertions
    • valgrind: always assertions are enabled, about 5% of time looks ok
    • nasa: most phases w/ assertions, landing assertions disabled
    • try to test w/ values sampled from entire domain
    • domain restrictions are acceptable as long as they are documented
    • for OSes we should test all values even if it's outside the specs (e.g. negative vals for >0 reqs.)
    • same for GUIs, the same trust boundary is present, which needs testing
  • on: 11 Nov 2012 - 1.23
    • use defensive coding, trying to prevent corner cases/errors
  • not on goal2013

Coursera - HPP

  • SM is dividing a thread block in 32 thread warps
  • 16x16 threads/block => 6 thread blocks => 1536 theads for Fermi
  • CUDA 3.0 supports up to 1024threads/SM
  • SIMD - same instruction on multiple data sets
  • control divergence when condition contains threadIdx
  • memory coalescing when a burst read is equal to a `tile' width => optimal memory bw utilization
    • corner turning - when copying from global to scratch memory, no penalty after end of copy using bursts
    • shared memory - SRAM - no need for bursting
  • arbitrary sizes for tiling of processing of the elements
  • operations need to be independent of each other (associative and commutative)
  • privatization - multiple threads write in output, use a reduction tree to merge the results in a single version
  • reduction computation - has identity value (e.g. 0 for sum, 1 for product) and is assoc. and commutative
  • avg. parallelism (N-1)/ log(N) for reduction tree - comparable to an efficient algo.

Coursera - Neural Networks for ML

  • enrolled_on: 03 oct 2012

Codecademy - Python track

  • started_on: 10 Oct 2012
  • on: 13 Oct 2012 - 3.1 / 8
  • on: 15 Oct 2012 - 4 / 8
    • universal imports - from module import <star>
    • function import - from module import function
    • import - import module
  • on: 1 Nov 2012 - 4.4 / 8
  • on: 6 Nov 2012 - 5.1 / 8
  • on: 12 Nov 2012 - 6.1 / 8
  • on: 23 Dec 2012 - 8.1 / 12
    • while/else, for/else construct - only when terminated w/o break
    • enumerate(array)
    • a, b, ... = zip(list1, list2, ...) - iterate over multiple lists at once
    • char in 'string' - correct syntax
  • list comprehension [ j**2 for j in range(10) if (j+1) % 2 == 0 ] # squares of odd numbers < 10
  • list slicing array[startendstride]
  • lambda functions
    • list_for_input=['1','2','3','4']
    • filter(lambda var: var == "3", list_for_input)
  • bin() to get binary representation of values
    • XOR (^) to flip bits using all 1's
  • class SUPER(): [...]
    • class SubClass(SUPER): [...]
    • override: super(SubClass, self).SUPER_method(params)

Books

  • Python Testing - Daniel Arbuckle
    • now_at: 103 (ch6)
  • Introduction to Neural Networks for Java - Jeff Heaton (http://www.jeffheaton.com/ai/)
    • started_on: 01 jan 2012
    • paused
  • Don't Make Me Think - Steve Kruger
    • started_on: 01 oct 2012
    • finished_on: 20 Nov 2012
  • The Lean Startup
    • goal: 2013

Finished

Coursera - Human Computer Interaction (HCI)

  • started_on: 01 jun 2012
  • now_at: 5.1 (week 3)
  • overall: 7/10
  • finished_on: 13 Oct 2012

Currently paused / postponed / cancelled

CS373 - Artificial Intelligence

  • started_on: 01 nov 2011
  • paused
  • goal: 2013