ML battle is heating up

The machine learning battle is heating up.

TPU card Deep Learning, the term, is somewhat recent and its popularity is most definitely increasing. At the root, the concept has been around for probably over 40 years under the umbrella or Artificial Neural Networks (ANNs). Although the idea was fantastic, the applications were limited by the availability of computing power. With the evolution of GPU cards (Graphics Processing Units), where each card packs thousands of core, deep learning is becoming more practical and applicable.

With their large scale data centers, Google and Amazon have both been able to polish their deep learning platforms. With everyone from the financial industry to healthcare now trying to research deep learning implementations for their own gains, both players are competing for attention.

After releasing to open source TensorFlow in November of last year, Google has take another step forward and announced TPUs or Tensor Processing Units. These custom ASICs were built specifically for machine learning and tailored for TensorFlow.

Meanwhile, Amazon is coming out with its own ML platform (DSSTNE, available on GitHub). You can read more on this at Geekwire.

Like with every new shiny toy, efforts tend to be overdone at first and we may see a few “Deep Learning” vending machines before things stabilize. And although it is hard for me to predict who will will the ML arms race, I think deep learning is definitely here to stay. It should soon graduate from “buzz” word to common computing practice.

A $1.5M Kickstarter Project Fails, Leaving Most Backers Without Their 3D Printer

Interesting news in TC:

Is this high profile and high value project failing to deliver a marketing and image blow to Kickstarter? Can it be interpreted as a sign that the bubble is starting to weaken and show its age. Or is this simply bad news about a single KS project?

Microsoft Is The New Google, Google Is The Old Microsoft

Very interesting article in Forbes on MS and Google:

I have a been a big Google fan and always knocked on Microsoft but they made interesting points. Google does spend a fair amount of time on less relevant projects though in my mind, it’s always been their approach.

They develop concepts and work on projects and see what sticks. Their focus is often purposely not on popular concepts. One reality remains, Adsense revenues are declining:

The knock-on effect: Adsense is declining and Google’s search market share is currently at its lowest point in seven years. Like Microsoft had done with Windows and Office, Google understandably still tightly holds onto the duo as its primary revenue pillars but the future implies only further slow decline with no obvious escape route.

Microsoft is trying to improve and making changes but they still on several important aspects. Their mobile game is irrelevant for one. Although they try to be an online and cloud provider, their offerings are often kludgy and integrate poorly with the desktop tools. Not to mention the annoying fragmentation in the Office 365 offerings (Personal, Small Business, Enterprise, etc with no transition path between the offerings).

Microsoft is a long way from being in a position to consider this ruthless second stage (or maybe not), but the first moves are well underway. The once ludicrous idea that Outlook and Windows Calendar could cause disruption on iOS or Android is now no laughing matter as Microsoft bought and effectively rebadged best-of-breed email and calendar apps Acompli and Sunrise.


Inside Chinas Surreal Housing Bubble on 60 Minutes

This is a great 60 minutes piece on China’s housing bubble. The middle class is turning towards real-estate for investment so builders are constantly putting up new buildings, developments and cities. But with only a privileged few buying multiple apartments, there is nobody to occupy these new developments who remain ghost towns for months and years.

As a side question: what would happen to the world markets if the Chinese economy was to implode?

7 Ways to Manage Email So It Doesnt Manage You | LinkedIn

An excellent article by Jeff Weiner at LinkedIn about managing email. Having done IT for several firms, I’ve seen it all. From people who simply can’t keep track, with Inboxes full of thousands of unread messages, to people who over process and over categorize, ending up with one category for each thread and spending 75% of their time managing emails.

Jeff presents great tips in his full article here.



Graph of Market Cap AnalysisI have recently discovered Quandl, which, per their website description:

[…] has indexed millions of time-series data from over 400 sources. All of Quandl’s datasets are open and free. You can download any Quandl dataset in any format that you want. You can also visualize, save, share, authenticate, validate, upload, index, merge and transform data. Our long-term goal is to make all the numerical data on the internet easy to find & easy to use.

Read More …

Unixrealm Posts 06/05/2013

Posted from Diigo. The rest of my favorite links are here.

Encrypted messages to kids

Crypto-WheelLast time I changed job, it meant an increase in commute (from 5 minutes to an hour) which also meant I would not see my kids every morning. So I picked up the habit (since then lost mind you) of setting the table for their breakfast and leaving them a note, wishing them a good day. Who knew that from a simple note, things would evolve to teaching encryption.
Of course, the simple note grew old over time so I started looking for alternatives. After trying the multi-color notes and various “analog” mediums, I turned to Bamboo Paper on the iPad and that provide several options. I also played around with a Boogie Board Rip and that provided an interesting alternative for a few days but it was short lived.
Read More …

Computational Investing, Part I at Coursera

I posted a little while ago about a Coursera class I took which covered financial analysis done with Python. The course was called Computational Investing, Part I. I find the topic interesting so I figured I would highlight what I enjoyed the most in a short series of posts.

The course, offered by Tucker Balch, an associate professor at Georgia Tech, covered various topics of portfolio management including several drawn from: Active Portfolio Management: A Quantitative Approach for Producing Superior Returns and Controlling Risk by Richard Grinold, Ronald Kahn.

Active Portfolio Management, by Grinold and Khan

We can breakdown the course in two distinct portions:

  • Lectures and theory :: The lectures covered a number of topics relating to instruments and portfolio valuations like:
    • Market mechanics
    • What is a company really worth
    • Capital Asset Pricing Model (CAPM)
    • Risk and Sharpe ratio
    • and more.
  • Practical work :: On the practical side, several homework was given where we explored computational techniques using Python, Pandas and QSTK:
    • Intro to Python/Pandas
    • Manipulating data with Numpy
    • Manipulating market data with QSTK
    • and more.

The lectures were well adapted for a beginner audience. Someone who understands the concepts could easily skip or fast forward through some of the lectures. I would sill consider the course useful as long as you are getting what you need from the practical homework.

Some students clearly did not have enough background with scripting and coding and struggled with the assignments. Yet they still benefited from the lectures and know more now about market mechanics, portfolios and risk.

Overall, I was very satisfied with the material presented in the course. Not perfect in any way but easy to adapt depending on your current situation and skill level. The lectures were interesting and provided some background information and more than enough leads for the curious mind to follow and research in depth.  One of the hardest thing to do when learning something new is motivate yourself to write useless code for the sole purpose of learning and practicing. This class provided the motivation required since the homework assignments forced me to code on deadline.

It is following the basic topics of the course that I embarked on my mission to dig deeper into Python, Numpy, QSTK and analysis of financial data using these tools.

In the next post, I will explore QSTK’s basic functionality.

Previous and related:

Next on topic (this is work in progress, most posts will show up in the weeks to come):

  • more to come…