My research in process mining

 When I rejoined the BPM team at QUT in 2014, many changes had taken place. The team had been relocated to the newly built P-Block. Arthur was still the leader of the team, but Marcello had been promoted to become QUT's youngest professor in computer science, managing a large group of human resources. Research within the team had switched its focus from process automation to process mining—a discipline created by Professor Wil van der Aalst, whom I have always respected as a mentor, friend, and rival. Moe was my direct supervisor, but frankly, I did not appreciate her research that much.

"Why do you want to return to BPM?" Arthur asked me during the job interview.

"I want to do open-source projects," I replied. "The SAIVT group where I worked previously does not allow me to share source code with others, which negatively impacts the promotion of my research outputs. Everything I did at SAIVT was sound, accurate, and objective; there is nothing I wish to hide. I want people to understand my publications fully. I admire your effort to host open-source projects and create websites for each small research area."

At the end of the interview, Arthur welcomed my return warmly. He mentioned that he had not received any notifications of bugs in C-YAWL, indicating it was developed to a very high quality. I was pleased to learn that YAWL had been installed on the Tianhe-II supercomputer in China.

Process mining is a research area that investigates event logs from information systems using AI techniques. In control flow, I view it as a machine learning area that focuses on data with timestamps as attributes. Within control flow, the core task is to automatically generate a workflow model from transactional logs stored in the databases of information systems. Typically, this workflow is represented as a Petri net.

Wil's team in the Netherlands operates a large open-source project called ProM. It uses a highly innovative method to incorporate new algorithms in this area as plug-ins, a strategy I consider to be very advanced. Wil's contributions to this field are by no doubt the largest in the world. However, at the beginning, I did not think he was exceptionally brilliant because, in my view, all the technical foundations were based on a single model: the Petri net. Later, I reflected on this and realized that in speech recognition, the primary model used was always the Hidden Markov Model (HMM) and its variations. We should not expect every model to be applied across a number of domains of applications. Artem, a young but excellent researcher from Ukraine, told me he heard Wil worked 12 hours a day. Then if that is the case, Wil is still very common! I thought to myself, though I did not dare to speak this out publicly.

The first time I felt Wil was truly unusual was during the writing of a paper on 3D visualization. I wrote the literature review section, and Arthur praised me extensively, saying it was exactly the correct style for a literature review, though he provided a few suggestions. While I agreed with him, I also thought his suggestions were merely theoretical, as I had never read a paper written in that manner. Wil later rewrote the final version. Ah! I fully admitted that Wil's version was much better. This made my own contribution to the paper feel minor.

However, Arthur suggested that I should be the first author, and he noted that Erik had agreed with this arrangement. I kept silent, and from then on I understood why the authorship of the C-YAWL paper would have problems. Then, Moe snatched the first author position. Arthur was confused and asked me why I did not compete for it.

"It's Erik's business," I told him. "Only he is qualified to compete for first authorship."

One day, Moe instructed me to draft the evaluation session for a paper on resource queueing, stating that we needed to finish it that same day. I worked incredibly hard but still could not finish it on time. I ultimately submitted the paper to them the next day.

Soon after, the annual performance review arrived just before the Christmas holidays. Moe blamed me for not finishing the paper on time.

"But I only delayed for a day," I argued.

"You delayed for a day, but Wil went on holiday immediately after," Arthur said. "Your one-day delay pushed back the submission of the paper by at least two months."

After the New Year, the paper underwent a few revisions before final submission. When Wil read the paper, he sent us an email saying that he liked it a lot, especially the evaluation section. Consequently, my draft remained unchanged in the final published version.

I was deeply grateful for Wil's confirmation. From then on, I believed Wil was truly a better leader than Arthur, even though I remained unsatisfied with how the C-YAWL paper was handled.

MAY SPRING COME

A few days ago, when I was jogging in Bernie Brae Park nearby, I was startled by a swooping magpie, which has allowed me to suddenly realise that it was already September and winter has quietly given way to spring here in Brisbane. As is my yearly springtime tradition, I played the Japanese song "May Spring Come”, and this year on the dizi (Chinese bamboo flute).

From a very young age, I have read countless poems and prose pieces praising the beauty of spring. Ironically, spring used to be my least favorite season. Growing up in my hometown, I remembered spring as being constantly wet, misty, and drizzly. The dampness often made me feel gloomy. "This is so different from the scenes described in poems!" I used to wonder. 

Later, I moved to Xi'an for university. Winter in Xi'an is bitter cold, causing all the trees and bushes—except for the pine trees—to lose their leaves. But when spring arrives, the foliage sprouts anew, bursting into a green so vibrant that you can hardly spot a trace of yellow or brown. Blossoms bloom, flowers flourish, and birds sing around the city.

Spring is always the best season in Brisbane. The days grow longer than the nights. The sunshine is bright but not yet too harsh. If you can dodge the swooping magpies, it is the most pleasant time to play on the park lawns, admire the colorful flowers, and enjoy picnics under the shade of the trees.

However, inside the house, you sometimes start spotting insects, ants, and cockroaches. The most annoying nuisance of all is the sheer number of houseflies. I remember when I was living with my grandma in Indooroopilly, many a morning a bowl of noodles that was prepared by grandma had already been waiting for me before I even got out of bed.

"I’ve cooked some noodles for you!" Grandma said.

"But wait!" she suddenly exclaimed.

She quickly grabbed a fly swatter and smacked a housefly resting on the edge of the bowl.

"Ah, I got rid of the fly!" she said proudly.

Uh, I really wish you hadn't done that! I thought to myself. I felt sick to my stomach, imagining the dirt from the swatter falling straight into my breakfast. But unwilling  to voice my thoughts to her, I ate the whole bowl anyway, pretending I didn't mind it at all.

---

From video to Wi-Fi sensing

Ten years ago, one of my classmates, Kien, who came from Vietnam and later worked at QUT, invited me to help him support his student’s project on multi-spectrum survellance event detection. This was because I had undertaken a student project on video surveillance event detection when we studied together on the same course. The term "multi-spectrum" means the integration of cameras, infrared sensors, and radar, as visible light, infrared light, and microwaves are all electromagnetic waves in different spectrums.

"Are you kidding?" I said to Kien. "Do you believe your student can understand that much within a course? Radar is a fierce animal that is hard to tame!"

"Well, we may not have time to do radar. But we've discussed starting with the camera, and using your previous project as the baseline. However, we need to use deep learning techniques..." he said.

"But I don't think the datasets in this area are big enough for deep learning models, which contain too many layers and parameters. I'm sure your model will be overfitted," I said to him frankly.

"At first, we will just use a model with pre-trained parameters," he said.

"You can do things this way if you merely want to have good experimental results to publish papers by using an overfitted model."

I said, "But the results are only beautiful on paper, and cannot be used in applications. Ah, there are too many papers on this topic without any quantitative evaluations. Those even include those published in top journals from prestigious universities like MIT."

I continued, "When I was a student, my supervisor, Simon, asked his supervisor Sridha whether we could do the project without evaluations. He said MIT's papers didn't have any experimental results but they still could be published in IEEE-T-PAMI. But it was said Sridha couldn't accept anything without quantitative evaluations, no matter how others did it and who they were."

Then Kien's boss, Professor Clinton Fookes, came to greet me. He said the motivation for him to apply radar was to tackle occlusions in visual cameras. "There are some papers with algorithms that can detect an event outside a wall!"

I was shocked to see Clinton so confident in radar technology, as I had never heard of him investigating electromagnetic waves. "Nevertheless, I'm interested in it," I told Clinton, adding that I liked physical subjects such as electromagnetism and antenna theory very much when I was in my undergraduate course.

A few years later, I found the student had graduated with a master's degree, while not making much progress in either radar sensing or deep learning.

During those years, I also spent a couple of years in Shenzhen working on a project about Wi-Fi sensing. No outputs were achieved when I finished the job. I then returned to Brisbane.

In my opinion, this area is still too challenging to have any sensible research outputs. I don't believe the results in any paper in this field, as there are too many fundamental settings that haven't been solved.

Currently, classical machine learning models are primarily for text, images, and audio. As all those models are constructed based on assumptions, and because the characteristics of different data media differ significantly, the models cannot be shared without a high degree of adaptation. In fact, there is still a lack of an effective model for surveillance videos. I don't believe existing models can be directly applied to communication channel data.

The methods to collect data, annotate ground truths, and evaluate performance are far from being solved. Researchers often highlight the motivation of tackling occlusions, but no one in the literature I've read highlights the discussion about how to avoid influences from outside the room.

When I was studying in Xi'an, our university had some labs for electromagnetic wave research. I often heard people blaming the previous university chancellor Professor Liang Changhong for disagreeing with the building of an overpass near the campus.

This was said to lead to a poor relationship with the local authority in Xi'an City. But I think there must be a reason for the university leadership team to object to the idea. Most likely, the vehicles on the overpass would interfere with the experimental results of the labs.

Abstract Painting

 In my opinion, painting is a vital visual method of communication. Sometimes, it is easier to express an idea through a degree of abstraction. Picasso had a famous lesson demonstrating the concept of abstract painting by drawing a bull. He showed that by removing excessive detail, a bull could be drawn with just a few strokes while people could still recognize what it was. One question we may ask is: why do we need to simplify it? I think it is because, in many situations, we want to highlight our focus by simplifying or omitting unrelated details. This is similar to writing an article; not everything should be described at length. If you talk too much about less important details, readers will find it hard to understand your main point.

In some situations, we may have to distort real scenes or add additional content to reality. This should always serve the requirement of expressing an idea. In Van Gogh's "Starry Night," the full picture is filled with stripe-like patterns. This, to some extent, reflects the painter's inner feelings. Despite this, we still know he was painting a starry night without others telling us.

I think to judge whether an abstract painting is good or not, we should have two criteria. First, it should be easy to understand. Second, the result of the abstraction should allow viewers to better understand the painting.

Recreation of Art From Viewers

Yesterday, I worked in the Brisbane CBD. After work, I walked into a store for art supplies. Unexpectedly, I found a few rolls of rice paper, which are typically used for traditional Chinese painting. I bought one and couldn't help but paint a picture on a piece of paper cut from it when I arrived home.

This was my first time engaging in traditional Chinese ink and wash painting since I gave it up twenty-eight years ago. Long, long ago, Chinese expression painting was my favorite form of fine art, and Qi Baishi was my most admired painter. In my childhood, I often followed his style to paint prawns, frogs, flowers, and fish. This time, when trying it again after twenty-eight years, I chose to paint a prawn. In only a few minutes, I had painted a big prawn on the paper.

I presented the picture to my wife and parents. They could understand what I was painting at a glance and said it was not bad for a first attempt in a long time.

I uploaded the picture online. When a French friend saw it, he left a message saying that it looked like a prawn that had been fed with contraceptive medicine. It is a known issue in agriculture that some people in certain areas of the world use contraceptive medicine to feed prawns, causing much concern over food security.

I checked my painting again and quite agreed with him that the body of the prawn was a bit too fat. There are many reasons for a prawn to grow fat. Not only those raised by farmers but also some from the wild are fat or have a similar appearance. Any of these interpretations is rational.

There is often a recreation process when admiring an artwork. Everyone has their own life experiences and perspectives, which are also always different from the creator's, no matter how similar they may be.

An excellent artwork should allow viewers to understand the most direct information shared with everyone, while also allowing each person to have a unique interpretation at a deeper layer.

This is the charm of any form of art.

My Study of Speech Signal Processing

 Many of my friends know that I have previously studied computer vision, communication channel sensing, and process mining, whereas very few of them know that my journey into artificial intelligence actually started with speech processing.

Near the end of 2007, when I was still an undergraduate student in Xi'an, I enrolled in an elective subject called "Speech Signal Processing," which taught the basics of speech coding and speech recognition. I have forgotten much of the speech coding material, but for speech recognition, I learned a very significant concept called a "instantiated stationary process." This concept set the foundation for using sliding windows in speech processing; within a small temporal window, a Fourier transform can be applied, which is known as the "Short-Time Fourier Transform." Then, the teacher introduced us to a mathematical model called the "Hidden Markov Model," which was probably the first model I ever studied in machine learning.

In 2009, when I started a master's degree course at QUT in Australia, I undertook a research project in the then Microsoft eResearch Centre with Paul. The team led by Paul was conducting analysis on animal sounds to understand the trends of eco-systems in Queensland. Though it was not for human speech, I gained valuable experience in audio processing.

Despite the fact that I later mainly conducted research in computer vision, the research group I was in still had some researchers working on speech and speaker recognition, and Sridha was the best among them.

One day, I discussed a problem with Sridha that I encountered in Gaussian Mixture Models (GMMs). It was about the covariance matrices becoming singular after some iterations during the training process.
"Yes, GMMs have this issue. This is a known limitation of the model, so I don't think it is your problem," he said.
"So you mean I shouldn't use it?" I asked.
"You can use it," Sridha said, “Every model in this area (machine learning) has its drawbacks, but there are always methods to solve them and apply them to our applications."
"I heard that Ahilan also used this model in speaker recognition. Is it because a speech signal is essentially a one-dimensional signal, and thus the covariance matrix is simpler?" I asked.
"No. Features for speech can also be high-dimensional vectors. There are many ways to deal with it," he said.
I then explained my thoughts on how to tackle it to him. "Do you believe it will work?" I asked.
"You can try," he said. "In our area, only experimental results can tell you if a method is working or not. If it is a method you thought of, you can write a paper if it works. If you fail in the experiments, at least you know it won't work."

Of course, the experimental results failed; otherwise, I would have written a paper on it. I knew my very good friend Ahilan was using a so-called "universal background model" based on GMMs for speaker recognition. That was why I often discussed his project with him. Soon, I found that the GMMs were trained on unlabeled data, and the datasets were massive. At that time, my understanding of machine learning had become much better. I had learned the concept of overfitting—a problem that occurs when a model has too many parameters to be trained on a small dataset. Then I realized that for GMMs, the majority of the parameters reside in the covariance matrices. This led me to think that what I had encountered was, in fact, a form of overfitting problem. For Ahilan's speaker recognition project, the labeled dataset was small, making the GMMs hard to train. However, a large scale of unlabeled data could be used to train the parameters to some extent. Then, there was a step called "enrollment" to finally adapt the model's parameters to the labeled data. I merrily went to chat with Ahilan about this, feeling as if I had discovered a treasure box on a remote island.

In 2014, I worked for half a year at a company where part of my work involved the development of speech recognition. I gained a lot of knowledge through this project. I found that accents in the training data significantly influenced the performance of the applications.

Recently, my father complained that Baidu Translate kept failing to translate his speech into English.
I said, "That's because your Mandarin is not standard or accurate."

A BEAUTIFUL WEATHER IS ALWAYS BAD

 Last Thursday, when I was on my way to the bus stop for work, it started to drizzle. I regretted forgetting to bring an umbrella, worrying that the rain might get heavier. "The weather is cunning!" I complained, as it had been sunny when I left home. What was worse, the rain suddenly became much heavier while I was waiting for the traffic light to cross the street. By the time I reached the bus stop, the rain had nearly stopped. I raised my head and, to my surprise, saw a splendid rainbow stretching across the sky over Gympie Road. I couldn't help but turn on my mobile phone to capture a photo.

The other day, I woke up early in the morning. As usual, I started reading my e-book. I was reading a biography of the renowned British painter J.W.M. Turner. There was a chapter about Turner's travels to the European continent. In his mind, beautiful weather was always bad weather. From the perspective of an artist, he was always keen to pursue scenes that couldn't be seen in normal situations. Extreme or bad weather gave him a lot of insight for his artwork.

Since I picked up my paintbrush again, I no longer complain about the challenges and suffering of the past. They suddenly became my treasures, as they gave me a lot of inspiration to create art.

Thinking about the past, I have experienced too many unexpected incidents. The more I suffered, the more I understood life. Especially whenever I was in a dilemma, I received support from family, friends, or even strangers. In addition to my own endeavors, I have overcome challenges one after another. If I can apply what I have learned or been inspired by into my artworks, I will be able to fulfill them—giving them souls that can touch many people's hearts.