Machine Learning Makes Finding Web Images Easier

Writing for the New York Times Science Desk, John Markoff reports on how computer vision and machine learning will create the next generation Internet where search engines find images and videos with the same degree of relevance as they do now with text.

And the need is crushing… in the next 60 seconds, YouTube will have uploaded 72 hours of video.

Today, unless images and videos are labeled, search engines have no way to match them against your query.  Even then, labels can be unreliable (e.g., “junk” versus the objects that comprise it).

To give search engines something akin to human sight, Stanford’s Dr. Fei Fei Li has teamed up with fellow computer scientists at Princeton to develop ImageNet, the world’s biggest image database.
Courtesy of

Given the enormity of the task and limited budget, Dr. Li connected with Mechanical Turk, the crowdsourcing system where, for a small payment per task, humans label photos. The database now has over 14 million images in over 21,000 categories thanks to the efforts of nearly 30,000 participants a year.

As the database of labeled images grows, machine learning algorithms enable software to recognize similar, unlabeled images. Over time, accuracy rates improve dramatically.

Surprisingly, when tested on a large collection of labeled images by Google computer scientists Andrew Ng and Jeff Dean, the system nearly doubled the accuracy of previous neural network algorithms designed to model human thought processes.

To further improve speed and accuracy, images are classified against WordNet, a hierarchical database of English words. With skillful programming to make educated choices about how to search the hierarchy, the database continues to rise to this growing challenge.

See the full article here.


Machine Learning Creates Win-Win for Lenders and Consumers

TechCrunch’s Leena Rao reports on ZestFinance, a startup poised to transform the payday loan industry with a new underwriting model that yields more accurate assessments of a borrower’s creditworthiness.

The result? Greater access to credit for more people at more affordable rates and fewer defaults.

Founded by former Google CIO, Douglas Merrill, ZestFinance weaves together traditional credit scoring with big data analysis, machine learning and human expertise.

From the lender’s point of view, the value proposition is the ability to better:

  • Quantify a borrower’s likelihood to repay
  • Manage the risks of their loan portfolios

So far, the model outperforms current industry best practice with a 54% lower default rate amid twice the approval rate for loans.

The human element comes in the form of ZestFinance’s team of predictive modelers who are experts in mathematics, computer science and physics.

As machine learning algorithms uncover thousands of variables, the team looks at them in the context of patterns and trends they are seeing before releasing them into multiple big data models that run in parallel.

In the process, load decisions are returned in minutes and ZestFinance continually improves its algorithms.

For more, see the article here.


Unsupervised Machine Learning Mines Useful Information from a Multilingual Sea of Text

The challenge of extracting meaningful information from an ever-growing Internet awash in languages, dialects, and knowledge domains is clearly, too much for our brains to handle.

Multiple Languages
Courtesy of

And traditional approaches are simply not up to the task.

However, a combination of statistical methods, data mining and machine learning could help change all that.

Mari-Sanna Paukkeri, a doctoral candidate at the Aalto University Department of Information and Computer Science in Finland has developed computational methods of text processing that are independent of any language or knowledge domain.

Languages share certain building blocks: symbols form words and words aggregate into sentences. Algorithms developed by Paukkeri analyze massive bodies of text and discover patterns to the presence of words and the structure of sentences. As a result, the meaning of specific words and sentences can be inferred.

To date, computational approaches to natural language processing have typically relied on rules defined in advance. Instead, Paukkeri’s algorithms use unsupervised machine learning to uncover meaning from statistical dependencies and structures that exist in the dataset with no help from data pre-processing or human intervention of any kind.

A familiar use of unsupervised machine learning and natural language processing would be the ability of Google News to bundle related new stories on any topic the user requests.

As such, Paukkeri’s methods have the potential to serve global corporations particularly well because they can glean meaningful insights from vast storehouses of data across multiple languages and knowledge domains.

Paukkeri has even studied how a search engine could ascertain if the user is an expert or a layperson and return suitable results by automatically assessing the difficulty of comprehension in the text it finds.

For more, see the original article here.


The Role of Machine Learning

Machine Learning is rapidly moving from classrooms and laboratories to every day use as applications that continually improve their performance become the norm.

This blog is dedicated to closely chronicling the spread of machine learning into business, engineering, medical, scientific, athletic, and entertainment software.

As such, we welcome contributions from interested people. Please send an email to with your expertise, ideas or leads on information about machine learning in connection with:

  • Making decisions and predictions
  • Performing tasks better as software gains exposure and experience with a given domain
  • Learning rules on the basis of background knowledge, examples and hypotheses
  • Discovering clusters in a large set of observations and figure out what’s meaningful with no prior knowledge
  • Modeling the way humans learn using neural network computing
  • Applying genetic and evolutionary principles to assess  the fitness of a population to perform a task
  • Predicting behavior in complex, real world scenarios over the short and long term

More specifically, feel free to share your interest in machine learning as it relates to:

  • Adaptive applications
  • Analysis of public sentiment
  • Autonomous robotics
  • Bio and chemical informatics
  • Computer vision, object and facial recognition
  • Data mining
  • Dynamic software development
  • Financial markets analysis
  • Gaming and simulation
  • Handwriting analysis and speech recognition
  • Machine perception
  • Medical diagnostics
  • Monitoring of structural integrity
  • Natural language processing
  • Pattern recognition
  • Recommender systems
  • Search