Sunday 15 May 2016

Response to the article: Three Trends That Will Define the Next Horizon in Legal Research by (FITZPATRICK, 2015).


Response to the article: Three Trends That Will Define the Next Horizon in Legal Research by (FITZPATRICK, 2015). In this article FitzPatrick is talking about the quintillion of data created everyday by humans, the decreasing storage space and the inability for people to handle this amount of data. He discusses tools such as natural language and machine learning as being able to assist professionals to bridge the gap of having big data and being able to dissect it,  he goes on to describe a group of people who occupied work force after the year 2000 referring to the as Millenials. His argument is that this group understand technology and are comfortable to use it better than those who started working prior to 2000. 

Natural language, machine learning and these Millenials is what he refers to as the three trends that will define the new horizon in legal research. He argues that machine learning will be able to help lawyers get accurate answers faster by learning from the available databases and learning how users interact with data adjusting their algorithms to be more accurate when similar situations occur.

I agree with FitzPatrick on the trends especially machine learning having the capability to improve how data is manipulated. I also agree with him on these young employees that he refers to as millennials, they happen not to be frightened by technology, like exploring and inquisitive. 

(Yang, 2016)Explains the use of actuarial method of risk assessment that compares individual behaviour to a norm-based reference group. He argues that “Since machine learning algorithm can be very good at detecting hard to observe relationship between data, it may be possible to detect obscured association between certain variables in criminal case and particular legal outcomes” the argument that agrees with the one for FitzPatrick.

In conclusion I agree with him these technologies will not only change the horizon of legal research but of many industries that take advantage of them and big data.

Referenes:



FITZPATRICK, S. (2015). Three Trends That Will Define the Next Horizon in Legal Research. Information Today.

Yang, J. (2016). Digitalization of the Criminal Justice Procedure and Applying Big data Analytics in Rationalization of Criminal Sentencing. Journal of hongik law review, 419-448.

 

A response on press release titled: Avik Partners Unfurls Machine Learning Service to Optimize IT Operations by Mike Vizard.


In this release posted on the 6 October 2015 on ITBusinessEdge Mike Vizard announces the unveiling of Grok by Avik Partners. There are other press releases by Chris Talbot and San Clemente on the same day on the same topic.
In his release Vizard explain Grok as a "Saas application for managing IT environments that first identifies optimal patterns in an IT environment to better identify anomalies that adversely affect application performance and then, secondly, continues to learn about the environment as new IT resources are added", he however does not elaborate on how this is being achieved and who will be using it. On the other hand Talbot and Clemente on their version explains that the application has a combination of adaptive and automation to detect unusual behaviour. They further explain that the application is for use by companies that uses public or private cloud services. I think this information is important for companies who will want to use this application; with less information companies may be reluctant to adopt the application.
Another point that Mike is not quoting from the CEO of Avik is that of using Saas approach which both Talbot and Clemente are quoting. I find Mike’s release less informative and hiding much information that could assist companies in making decisions on whether to adopt Grok or not.
References:
Mike Vizard, 06 October 2015, http://www.itbusinessedge.com/blogs/it-unmasked/avik-partners-unfurls-machine-learning-service-to-optimize-it-operations.html

San Clemente, Calif. (PRWEB) October 06, 2015 http://www.prweb.com/releases/2015/10/prweb13004074.htm

 

Saturday 14 May 2016

Machine Learning Tools and how Companies Leverage on Machine Learning Algorithms


Image result for machine learning tools

As with any other technology we see companies taking advantage and building machine learning tools that they avail for use either on the cloud or can be dowloaded and used at your local machine.

At the for front we see IBM with Watson that offer varried servises, Microsoft with Azure and Predictive Analytics, Google with Google Translate and Google Prediction API and Amazon with Predictive Analytics with Amazon Web Services.

There is also a lots of startups and open source machine learning tools that are battling their space with Loius Dorard mentioning PredicSis and BigML as providing a competing API’s when compared with the top 4. Below is a list of some of the startups and open source tools:



How then; can companies leverage on this technology that is obviously gaining momentum. Daxx mention 6 ways that companies are leveraging Machine Learning Algorithms as:

1.     Price Optimization

2.     Improving Customer Engagement and Maximizing Profits

3.     Launching Targeted Promotions

4.     Predicting Equipment Failure

5.     Detecting and Preventing Fraud

6.     Streamlining Talent Acquisition

If your organisation has not thought about how to use machine learning this is the time.

 References:



Wednesday 11 May 2016

An Interview with Bennie Leonard a Machine Learning Scientist at DataProphet - Special Post

DataProphet is a South African, Cape Town based Consulting company that specialises in Machine Learning. I have had an opportunity to ask some questions from Bennie Leonard who is a Machine Learning Scientist at DataProphet. Thank you Bennie for the insights.



Q. I see you are a Machine Learning Scientist at DataProphet. Where did you study machine learning?

 A. I studied computer science at the University of Pretoria. I specialised in optimisation algorithms, with a focus on swarm intelligence. Optimisation algorithms are a class of clever search methods and are often used to enable machines to learn from experience.

Q. Why did you choose machine learning as a career?

 A. Programming computers to learn has been a field of interest for scientists for at least a few decades. However, over the last ten years or so, the field has gained substantial traction in real-world applications. Machine learning is widely used in the technology industry to perform a range of tasks, including product suggestions in online shopping, search prediction for online search engines, and even mastering difficult games like Go. Even so, there are still a huge number of businesses that are either unaware of the capabilities and potential benefits of machine learning, or struggle to understand how to apply machine learning to their unique business environments.

The enormous potential that machine learning and artificial intelligence has to offer, and the excitement of working in a very young and developing field, are what drove me into a career focussed on machine learning. At DataProphet we aim to understand and fill the gap between scientific advances in machine learning and the useful application thereof to individual businesses.

Q. What algorithms do you apply on your job as a machine learning specialist and why?

A. Different applications of machine learning often require unique combinations of algorithms to perform a given task. Our expertise at DataProphet ranges from relatively old (and commonly used) tree-based classification methods to the most recent developments in deep neural networks. Which specific algorithms to apply depends heavily on the scope and specifications of each individual project.

 Q. We have professionals that are well known for their work in machine learning, who do you look up to?

 A. There are many highly respected professionals in the field, but progress rarely comes without relying on the work of other scientists. Personally, I have deep admiration for the likes of Alan Turing, John von Neumann, and Ada Lovelace, who played crucial roles in laying the foundations for the science we build upon today.

 Q. What is the adoption rate of machine learning in South Africa?

 A. It’s hard to put a figure on the adoption rate of machine learning in South Africa. While it is definitely increasing, we often find that businesses are either overly optimistic, or overly skeptical when it comes to machine learning. There is still a lot to do in terms of educating people as to what the capabilities of machine learning are. With a better understanding of the technology, the adoption rate will likely increase faster.

Q. I had a presentation on machine learning where I gave an example of (FITZPATRICK,2015) article “Three Trends That Will Define the Next Horizon in Legal Research” where he talks about machine learning as one of the trends. The question that I got was around ethical issues when training these models. What is your take on that?

 A. As with any technology, it is important to consider the ethical implications. Machine learning models are trained on the data that humans provide them with. In that sense, the algorithms are general-purpose algorithms. They will attempt to understand any data that is presented to them and the trained models can be applied in any way we wish to apply them. Throughout history (and still today) there are many examples of technology being used in unethical ways. It would be naive to think that machine learning is somehow immune to this possibility. Indeed, companies like Google’s Deep Mind and the non-profit OpenAI have already established ethics boards to guard against the unethical application of artificial intelligence, and rightly so.


However, the intent of scientific research is ultimately to expand and enrich human knowledge. That intelligence forms part of who we are is indisputable. And in our quest to truly understand intelligence, we will undoubtedly learn more about ourselves. So we are faced with two choices: we can either continue on this path to discover the true nature of intelligence, while being mindful of the potentially far-reaching ethical implications of what we might learn; or we can credulously decide that the risk is too great and be willfully ignorant about this mysterious quality we call intelligence, that is such a big part of who we are. We should all choose the former.

Bennie Leonard

Monday 9 May 2016

Some Applications of Machine Learning

In this section I will discuss some machine learning algorithms and their application.  Will discuss supervised learning and unsupervised learning, active learning and transfer learning.

Supervised learning
In supervised learning the data is labelled, machine learning algorithm maps the input to the desired output to generate a model. This technique is commonly used to train neural networks and decision trees. Neural networks are mostly applied where there is previous data to learn from like in character recognition, image compression, stock market prediction etc.  While decision trees are also applied where there is previous data to learn from, they are mostly applied where there are decisions to be made like in product planning and loan applications.

Unsupervised learning
In unsupervised learning, machine learning algorithm draws inferences from datasets consisting of input data without labelled responses. This technique is based on data mining methods which include clustering and latent variable methods.   Some of the unsupervised learning applications are language understanding and image identification.

Active learning
Active learning is a semi supervised machine. In this technique, the learning algorithm is allowed to ask questions from an oracle.  The oracle is a human annotator which can assign labels to training instances, in return the learner uses the feedback to find or improve a model for the training data. They have been successfully applied in regression testing, fuzzy testing and inference of botnet protocols.

Transfer learning
Transfer learning system learns models with different “source” sampling distributions and training labels, and then transfers that knowledge to the target task (Perlich, Dalessandro, Raeder, Stitelman, & Provost, 2014). Transfer learning attempts to change this by developing methods to transfer knowledge learned in one or more source tasks and use it to improve learning in a related target task (Torrey & Shavlik, 2009). This technique has been successfully applied in online advertising.

Next we going to look at some available machine learning tools.

Reference:
  1. Aarts, F., Kuppens, H., Tretmans, J., & Vaan, F. (2014). Improving active Mealy machine learning for protocol conformance testing. Machine Learning, 189-224.
  2. Pete Swabey, February 24th 2014, from http://www.economistinsights.com/technology-innovation/opinion/%E2%80%9Cunsupervised-learning%E2%80%9D-and-future-analytics targeted display advertising: transfer learning in action. Machine Learning, 103-127.
  3. Torrey, L., & Shavlik, J. (2009). Transfer Learning. Handbook of Research on Machine Learning Applications.
  4. Unsupervised learning. (2016, March 30). In Wikipedia, The Free Encyclopedia. Retrieved 13:07, May 9, 2016, from https://en.wikipedia.org/w/index.php?title=Unsupervised_learning&oldid=712692172

Wednesday 27 April 2016

Ethics in Machine Learning - Special Post

Can't help but to mention that as I presented on the topic on Monday 25 April, a question of ethics was one that seriously kept me thinking.  So as I was reading I came across this link: http://www.kdnuggets.com/2016/03/ethics-machine-learning-tay-chatbot-fiasco.html

I think its a great initiative, one that I would like to follow closely.



New Era of Machine Learning



Machine has definitely shifted from the times it was based on theory, a thing people were talking about in the corridors before yet another conference of machine learning.  More and more applications of machine learning are now reported.

We have seen technology leaders like Google, Facebook, Microsoft; even banking industries implement these powerful technologies. 

We are now leaving in the world that is driven by technology; everyone talks about big data, cloud computing, adaptive security to mention few and Machine Learning seems to be the heart of them all. Without Machine learning it would be difficult to handle these terabytes of data and put defensive mechanisms against the ever improving attackers.
 
So in this era Machine Learning is revolutionizing the world we live in. Of great importance though is to mention that machine learning is still based on the very algorithms that were founded in the 80’s, which makes it a subject that is still very much dominated by academic specialist and researchers. We see Google hiring the likes of Sebastian Thrun, Fernando Pereira, Ray Kurzweil, all academics from different Universities. Facebook hiring Professor Yann LeCun of NYU, and Baidu which is considered to be China’s google hiring professor Andrew Ng from Stanford who previously worked at Google.  The completion gets tighter in this space.

 “If you want to beat the crowd now, you have to try and buy the people that really know this stuff—otherwise you’ll be a few years behind,” by Michael Mozer, from Colorado University.
Let’s look forward to discussing some of the applications of Machine Learning.
References: