Author: abagiyev

There are some questions that I believe can have a positive impact on the successful implementation of the project in the name of time, accuracy, and scope.

1. As it is always discussed how much data is important. So, in our case, there are some publicly available data already but to have the model tested in the real environment, the creation of the custom dataset is one of the goals. I would like to hear about the successes and failures you have had in your own projects and what was the reasons, whether it is a common pattern or depending on the scope and content of the project.
2. The second question can be about the quality of the data. How to successfully detect the format of the data required to gather such as if video whether it should be mp4, Flv, MOV or totally another one. So, some formats are taking less storage but also carry less information. Is there any pre-defined method to find detect the one without trial/fail period?
3. How to find/decide on the correct metrics to evaluate the success rate of the project? Let's say it is up to us and no input has been given by the stakeholders. What strategy do you follow?
4. Let's assume the model is built and working fine about some average and there is nothing that seems to be wrong. How to debug the system to see whether there is any mathematical mistake in the computation which just ended up not giving so bad but in the long run it would fail? What methods do you follow to be sure that the model is correct on the mathematical base?

During the initial briefing presentation, there were a lot of points to gather our attention for further investigation and possible methods and tools are one of them. In this blog, this has been my prior concern to focus on.

Group-based emotion recognition (GER) is an interesting topic in both security and social area. In our work, by utilizing the Neural Network, emotion recognition (ER) can be performed from a group of people. Initially, original video frames are taken as input and pre-process from multi-user video data. From this pre-processed image, the feature extraction is done by Multivariate Local Texture Pattern (MLTP), gray-level co-occurrence matrix (GLCM), and Local Energy based Shape Histogram (LESH). After extracting the features, certain features will be selected using the Modified Sea-lion optimization algorithm process. Finally, a recurrent fuzzy neural network (RFNN) classifier-based Social Ski-Driver (SSD) optimization algorithm is proposed for the classification process, SSD is used for updating the weights in the RFNN. Python platform is utilized to implement this work and the performance of accuracy, sensitivity, specificity, recall, and precision is evaluated with some existing techniques.

The above-named methods are one of the many possible solutions to the problem domain. They just become more suitable due to the high accuracy in the final result. Still, there is a need for further investigations on the yet unrevealed ones and to compare the possible cons and pros.

In data science data means more than one thing. Strong data is a powerful start, a solid way to success, and meaningful results. It is easy to form a sentence to express data but in reality, this is like finding Alice in Wonderland, because great data is very rare to find and it is divergent, not restricted to certain rules, but still there are some pre-defined metrics to evaluate. It is lucky that neither I nor my teammate will be the first person to establish the path. There are many pre-built datasets out on the internet to be used by the public and Kaggle is one of the places to rely on.

However, there are some concerns to take care of while deciding on the dataset(we believe finding the correct dataset can guide us to the correct project definition that we would find a joy to carry on). It has been disclosed that all master projects in one or more ways have to be tied to local business needs and therefore, this should be the primary filter on finding the correct one. The latter one is set by ourselves, which indeed is having a project related to computer vision(literally, there was a third one as well about having the project related to medicine, but due to the reality of not finding appropriate business needs retained us from that idea). Currently, it is the examination period of the publicly available datasets to selects some of them. After having all the remaining choices, we will try to group them up on the reliability, relativity, and impact rates. Basically what it means is that the data must be relevant to Azerbaijan(for example, if doing sign language for Azerbaijani, we cannot choose an Indian sign language dataset), the dataset should have enough qualified data, and previously done researches on that data should reflect over 75 success rate.

It is obvious that since the start of the summer term, my approaches and thoughts about choosing a master project have changed positively. However, I believe that still there is a contradiction between ADA requirements and GWU project suggestions. In the classes, we spend time learning how to find some interesting project which will solve a globally existing problem and it is better not to have any repetition on this(for example, self-driving car project is not advisable because there are some people out there who has already done a great deal on this) but in term of ADA requirements, the idea should be bound to local issue and no need it to be something unique(maybe, it has already been done but not in this particular sphere). This contradiction still drawbacks me from making solid decisions and researches on finding the most skillful project that would force my boundaries when there is no guarantee that it will get accepted and rather to wait for the other project suggestion from the ADA side.

It can be said that the classes are helping to improve our mindset to comprehend the problems completely. The told real-life stories increase the courage on trying new things because usually, humans are afraid of making mistakes, and seeing that it is actually part of the professional development boosts the learning process.

I have always believed that success is built on failures if the causes have been discovered and solved. Having class exercises is a great activity to push us toward making critical thinking on the given problem domain. This makes the first part of the failure story and the later part is completed by the individual feedback of professors, so we can find and solve the issues.

The classes are thought by slides which are very efficient to enable easy tracking of the instructor's speech. There are highlights and thinking required quotes to keep the audience engaged. The usage of the blackboard class page is well designed to support easy usage. Also, the professor is doing a great job of keeping students up to date about any significant change. All necessary resources such as books, articles, and tools are provided which significantly helps us to save time on researching and mainly focus on the subject.

I am Aydin Bagiyev, current master student at a joint program called "Computer Science and Data Analytics" offered by ADA & GW universities. Here, is the brief introduction of me.

I was born in Zaqatala, Azerbaijan and have travelled some amount of the countries such as Austria, Germany, USA, Spain, Lebanon, Turkey and some others. This is my second visit to USA and in the previous one I was just 16 and the destination was California state.

Within Computer Science, I am quite interested in Computer Vision and Deep learning. In future, I would like to syntesis my computer science knowledge with robotics field.

My special interest in this field is toward medical image understanding and aerial detection.

Outside of Computer Science, I am interested in soccer, board games and exercising. My favorite book is "Oblomov" by Ivan Goncharov.