Methods for Unsupervised Semantic Modelling (by Professor Wray Buntine)

(11/1/2016)

The seminar given by Prof Wray shared some non-parametric methods for unsupervised modelling based on an example in NLP (natural language processing). With current growth of information systems, we need some better tools to help us to deal with the information overload. Such tools should be capable of organizing, searching, summarizng and even understanding the information. The 'understanding of the meaning of natural language' or in other word 'semantic' is the main focus of Prof Wray's methods.

The unsupervised learning in this model is based on Dirichlet distribution from probability and statistic. Throughout the presentation, Prof Wray tried to avoid the complex maths functions he has used. Basically, the concept involves probability vectors for the following elements quoted from Prof Wray's note:

• the next word given (n − 1) previous, 
• an author/conference/corporation to be linked to/from a webpage/patent/citation, 
• part-of-speech of a word in context, 
• hashtag in a tweet given the author.

A Dirirhlet distribution is then used to develop Dirichlet processes for the semantic model. It enable the approximation of vocabularies or documents hierarchically. The benefit of this model as per said by Prof Wray is that it reduces parameters optimisation problem faced by most distribution functions. Also, the nested (or hierarchical) Dirichlet processes have fast samplers as compared to others.

This is indeed a very high level of learning for me as the concepts and functions involved are really something new to me. Anyway, thanks to the seminar, I am more open to some new and advanced algorithms for my research project.

Communicating Research (Workshop 4)

This workshop is the last of the series of Communicating Research in 2015, and it focused on profiling our research. Research profile, when constructed with the right amount of details, would become a basic tool to promote our research across different digital media.

In this workshop, we were told about some common social media that we may consider for promoting our profile. Examples of academic/research social media are Research Gate, Acedmia.edu and Google Scholar. The non-academic profiling media could be LinkedIn or even Facebook. I have personally joined Research Gate a few months ago, in which I find that it is quite a good tool in the way that it tries to connect you with other researchers when common research areas are found. But I don't really like to use Facebook or or other non-academic social media for the similar purpose as I feel that people within my Facebook contact are generally not researchers. However, after some thoughts shared by Julie Holden, I do agree that other social media like Facebook, if organized and managed properly, can be a good place to promote our research as well.

Knowing the benefit of profiling our research, especially with the help of social media, I begin to fill-in more details in Research Gate and start organizing my Facebook contents. Hopefully, as suggested by this workshop, our profile will benefit us not only in our career, but also gets our findings and works reaching out to more people, inclusive of even friends and family members who aren't experts in our field. One day, they are going to understand and appreciate our works.

PS: At the end of the workshop, we were asked to complete our profile based on a template provided to us, and with delight, the profile is going to be published at Monash website! That's great.

Communicating Research (Workshop 5)

This workshop focused mainly on securing of funds and grants for research projects. There are several important things that we as the researchers have to be cautious of. The following sections highlight some of the learning points from the workshop.

About title of the proposal:

Clearer, bigger picture, easily understood

Structure of proposal:

• summary
• introduction
• literature review
• approach

Proposal omits results and discussion. Some planning and capacity such as budget, track record are needed. A discussion using the analogy of wheels was given as follows:

• big picture (need for efficient locomotion)
• the question (can the wheel provide efficient locomotion)
• paradigm (proof of concept, experiments, metrics, benchmarks)
• specific goals
• approach

Sometimes it is useful to get the best structure from question to answer:

Step 1: interpret the question

Step 2: subgoals

Step 3: choose methodology

Formula to writing summaries:

• Background (why is it significant)
• Question (what are you doing)
• Approach (how are you doing it)
• Outcome (where the result will lead to / where will it get us)

After all, it is the funding organizations that will be the main readers at the point of the fund application. Hence, it is the prime concern to make sure they (which could normally be non-expert in our field) are presented with easily understood contents.

Milestone Workshop by Prof Sue McKemmish + Pecha Kucha (Presentation of local HDR students' projects)

In conjunction with Prof Sue's visit to Sunway campus, we had a chance to be briefed on the PhD milestone, and to meet up with Ms Cassandra the Graduate Student Services manager. The workshop is pretty much self-explanatory based on the following diagram. Some questions were raised regarding the courses (FIT5144 & FIT6021) compulsory to the students, and how would the timeline be affected with the inclusion of these courses. We were told that an extra period of 3 months is eligible for those taking the courses - if extension is needed. This has made most of us relief as the works from those courses have taken up quite some significant time out of our research activities.

In another session named as Pecha Kucha (a presentation style based on 20 slides per 20 seconds) was the students' presentation of their research projects. Each of the students was given 20 seconds to present the current stage of their research to the audience, which was inclusive of most of the HDR students, some lecturers from the school as well as the Deputy Head of School, Prof Sue herself, and Ms Cassandra. Based on my presentation, I received an important feedback from Prof Sue, in which she suggested I could consider to include my end-user (the Blind and Low Vision) along the prototype development phases. The point was taken and now it has formed part of my research methods. I appreciate the session.

Overall, the meeting and the workshop has offered us a platform to know more about some policies and procedures important to our PhD milestone.

Series of Mathworks' Webinars

Based on my previous positive experience with Mathworks online seminar series, I joined other 3 seminars from 13th August to 1st September. I summarized them in the following sections.

Integrating MATLAB into your C/C++ Product Development Workflow (13th August 2015)
As I will be coding my prototype mainly using C/C++ in Robot Operating System and Linux (specifically Ubuntu), I found the highlights of this seminar are definitely helpful. At the early stage of my research, I prefer to use Matlab for prototyping. However, due to the needs of implementing the actual working application on an embedded platform for my wearable technology, I have to eventually switch from Matlab code to C/C++. So this seminar has shed some lights on the integration of my Matlab code into the development workflow. Technically, I can tackle the issue from both ways - either by generating C code from Matlab code to seed new designs; or integrating my existing C code into Matlab for simulation or prototyping. These could have made things much more faster and easier for me.


PID Control Made Easy (18th August 2015)
A proportional–integral–derivative (PID) controller is a control feedback mechanism commonly used in industrial control processes or systems. I have enrolled into this seminar mainly due to the belief that I might eventually be using some sort of control mechanism in the prototype to provide feedback to the blind and low vision users. This is something I am going to implement at the very end of the prototype development. The seminar introduced Matlab Simulink Control Design toolbox as a straightforward and partly automated process to tune PID controllers. The environment also allows automatic code generation to deploy PID controllers. I am seeing that the techniques learned here might be relevant in designing my HCI feedback control.

Signal Processing and Machine Learning Techniques for Sensor Data Analytics (1st September 2015)
This is the most relevant topic for me among all the series from Mathworks seminars. My project application needs the joint use of signal (image) processing and machine learning techniques on real-time and sensor data. Based on the speaker, MATLAB can speedup the development of data analysis and sensor processing systems. This is achieved by having a full range of modelling and design capabilities within a single environment in Matlab. In the seminar, an example form a classification system for human activity is shared and discussed.

Having acquired some skills from these seminars, I am looking forward to putting them into good use of my research project.

Image Processing Webinar

(23/7/2015)

So this is my second series of the Mathworks webinar. This time, we have had Brett Shoelson (a PhD and the Principal Application Engineer for Mathworks, specialized in image processing) to conduct the seminar on image processing by using an example form a research work in Finding Parasitic Infections with Matlab. 

Somehow the example from this seminar is more relevant in biological science (Parasitic Infections), however, the algorithm and techniques involved are absolutely relevant to computing and IT. We were introduced to some typical real-time challenges in the discipline, and some discussions about the possible approaches in dealing with them.

The part that I am really benefiting from this seminar is the techniques from computer vision and machine learning, which were used to automatically recognize (quantify as per said by the speaker) the target in the image set. In my own project, I will be focusing on computer vision to acquire real-time ground data, and this data will be then processed by some means of image processing probably involving machine learning. I downloaded the example codes and hopefully they will be pointing me to some relevant solutions for my project.

A Webinar on Data Analytics with MATLAB

(17/7/2015)

I am an experienced user of Matlab - a high-level language and interactive environment for rapid scripting and algorithm testing (at least that was how I have used Matlab all the while for the past 10 years).

A few weeks ago, I was notified about a webinar from Mathworks (the company that developed Matlab) on the topic of Data Analysis. The advertisement caught my attention with the following short description: "Using Data Analytics to turn large volumes of complex data into actionable information can help you improve engineering design and decision-making processes... and so on"

So I signed-up the seminar, and wow! I have actually learned some new and interesting techniques from it.  It was a short 1-hour session, but the pain is the time difference between my local zone and the organizer's zone in the United State (for me that was some time in the midnight).

Let's get back to the topic, so I have got a chance to learn techniques for exploring, visualizing and combining complex multivariate data sets, which I don't normally perform using Matlab. Another technique which could be closer to my current research project is prototyping and testing predictive models using machine learning methods. This is something advance and new to me, and to pick up the skill is luckily not that hard. To conclude, this was a really useful online seminar, at least technically.

PS: Following this, I have signed-up for more webinars from Mathworks. More to come in my next post.