Full Moon Crescent Saber: Prologue

Tonight was a night of a beautiful full moon, so I thought it would be the perfect time for me to start this translation project, something I've wanted to do for a long time. :)

Full Moon Crescent Saber is a book started by Gu Long and finished by Sima Ziyan. Because of that, it is also the most controversial book of Gu Long. I like this book because of the unique artistic conception and atmosphere described in the book, which really made it stand out from all other Gu Long books.

============================================================

Full Moon Crescent Saber
-- Written by Gu Long, Translated by Lanny Lin


-->

Prologue

Full Moon

The moon may be wax or wane. The story we are telling here is about the full moon, because it happened at a night of the full moon. At that night, the moon was more beautiful than any other night, with a magnificence so mysterious, so bleak, and so heart-breaking.

Same goes with the story we are telling, a story filled with charms so mysterious yet beautiful and fantasies so mystifying yet stunning. As told in the ancient mythical tale, when the moon rises in the dark nights, there are always fairies dancing in the moonlight – fairies of the flowers, fairies of the gems and sapphires. Even dark souls and fairy foxes living deep underground would come out to worship the full moon and to draw in the vigor of the bright moonlight.

Sometimes they will even transform into human forms, live in the human world as many different characters, and do things no one would ever imagine.

These things are sometimes startling, sometimes heartwarming, sometimes frightening, sometimes exhilarating, and some other times beyond imagination. They could rescue someone from the deepest abyss; they could also shove someone off the steepest cliff.

They could give you all the fame and fortune in the entire world; they could also make you lose everything you’ve ever own.

No one has ever seen their true faces, but no one could deny their existence either.

Crescent Saber

A saber may be straight or crescent. What we want to talk about here is a crescent saber, as curvy as Qing-Qing’s eyebrows.

The crescent saber belonged to Qing-Qing. Qing-Qing is a beautiful and mystic girl, just like the full moon of that night.

Sabers are weapons made to kill.

Same goes with Qing-Qing’s crescent saber. When the crescent reflection flashes by, calamity befalls; no one can escape the calamity, because no one can get away from the crescent shine of the saber.

The shine of the saber is not hasty; it is as smooth as the moonlight, but as soon as you see the moonlight, it has already befallen upon you.

There is only one moon in the sky; there is also only one crescent saber on earth.

It doesn’t always bring calamity when it appears in the mundane world. Sometimes it also brings people righteousness and happiness.

So when it appears in the world once again, what will it bring to this world this time?

No one knows.

Qing-Qing’s crescent saber is also emerald green^[1], as green as the verdant distant mountain, as green as the spring trees, and as green as tears in young lovers’ eyes.

On the emerald green and crescent-shaped blade is a line of tiny words, “All night in the attic I hear the spring sprinkle^[2].”

Fortunes may be as unpredictable as the winds and clouds in the sky. The moon may be dim or bright, wax or wane. Perfection in life was never easy to come by.

May we all be blessed with longevity. Though miles apart, our hearts still cross through the beautiful moon high in the sky.^[3]

---

[1] In Chinese, “Qing” means emerald green.

[2] A verse from a poem of Lu You (1125-1210 AD), a poet from the Song Dynasty. I will not attempt to translate the entire poem here.

[3] The last few verses are excerpts of a very famous Mid-Autumn Festival poem by Su Shi (1037-1101 AD), a poet from the Song Dynasty. Here’s my poor attempt at translating the poem:

Prelude to the Melody of Water

When did the bright moon first ever appear?

Raising my wine cup I ask the blue sky.

High above in the moon palaces,

Wonder what year it is tonight.

I want to ride the wind and fly to the moon,

But I fear the jade terrace is too cold and high.

I’d rather stay in the human world,

And dance with my shadow in the moonlight.

As the moon rounds the red pavilion and slants through the silk-pad windows,

It shines upon every wakeful eye.

Moon, are you bearing any grudges.

Why always the full moon when loved ones are not nearby?

People may have joy and sorrow, parting and reunion,

The moon may be dim or bright, wax or wane.

Perfection in life was never easy to come by.

May we all be blessed with longevity, though miles apart,

Our hearts still cross through the beautiful moon high in the sky.

Now support the translator Lanny by following my blog and leaving comments! :)

Videos of the Day:

The beautiful poem referenced above was turned into the lyrics of a beautiful song sang by Teresa Teng, a huge pop icon in the 70s and 80s in the last century. Enjoy!

While searching for Teresa Teng's video, I ran into the video below on YouTube and was very impressed by the talent shown. A girl used her own music composition for the same poem and showed her beautiful voice. Even if you don't understand a word of hers, you'd still enjoy it (someone left a comment saying exactly that)!

AI and Robots: BYU using computer vision to catch parking violators

In the past, faculty members, staffs, and students at BYU (Brigham Young University) had to obtain and place special stickers on the windshield of their cars every semester if they want to park on campus in designated parking lots. Starting in Fall 2009, thanks to the new computer vision technology adapted by BYU police, this step is no longer necessary.

There are four types of parking lots at BYU: Faculty and Staff Parking, Graduate Student Parking, Undergraduate Student parking, and Visitor Parking. Because faculty parking lots are everywhere on campus, while graduate student and undergraduate student parking lots are relatively further away from the center of the campus (graduate parking lots are slightly closer), many students are tempted to park at faculty parking lots just briefly for a class period of about one-hour. Many used to be able to get away from it because of limited parking officers, but that is probably coming to an end because campus police has a better weapon to fight parking violators.

An automatic license plate recognition system, developed in Israel (I suspect this company) and made its way into US through Canada (don't ask me why), has become a very powerful tool for BYU police to catch parking violators. Cameras installed on top of police cars (as shown in the picture on the left) can automatically take pictures of cars in the parking lots. License plates are recognized and matched against a database to quickly determine if the car can park at the parking lot. An alarm is played when a violator is identified, and with just a push of a button, a parking ticket is automatically generated and printed. Parking officers can now quickly drive around campus multiple times each day and get their job down all with the comfort of sitting in their seats.

The picture on the right shows a closer view of the type of camera in use. The same kind of camera is also used at gated areas to automatically raise the gate for eligible cars. The accompanying software can read 60 plates a second and can recognize a license plate on a car going 120 mph with the help of the high-speed camera and fast computer algorithms in recognizing numbers and letters. The system also a GPS built-in, so images of cars are also geo-tagged with GPS locations in case people forget where they parked their cars.

Inside the police car, a very durable tablet PC is mounted on the panel so the parking officer can interact with the software using a stylus pen. A wireless keyboard can also be used to enter license numbers into the system.

Obvious benefits of the system include: more efficient patrol of lots of parking lots, comfort of staying in the car in extreme weathers (hot or cold), and automatic alert for stolen vehicles. However, this technology also has its drawbacks. For example, in cases of heavy snow (which is not so rare in Utah), the license plate might be covered by snow and not visible. Also since the parking officers can now do most of their job without getting out of the car, special parking spots like the 15-minutes ones are getting less attention and could be abused more frequently. In the past, people who owned multiple vehicles had the option of hanging a badge in one of the cars. This also means only one car is allowed to be parked on campus because there is only one badge. With the new system, since there's no sticker and no badge, all cars can be parked on campus at the same time. Lastly, privacy is also a concern because now the campus police can easily identify when cars are parked where each day.

So how does the recognition work? There are two main challenges: 1. Identify the license plate in the picture. 2. Recognize the license number. I don't know the exact algorithms used in the system, but based on techniques learned form my computer vision class, I certainly can come up with some intelligent guesses. Identifying the license plate in a picture probably relied on edge detection techniques combined with detecting high-contrast areas that also have the rectangular or rhombus shape. A coarse to fine search is also likely. Recognizing the letters and numbers is relatively easy with machine learning classification algorithms such as decision tree or nearest neighbor.

It is worth mentioning that such license plate recognition systems are already widely used by police forces. The video below shows an example. If you live in California, then you probably have heard stories where people get their traffic ticket in the mail together with a picture of their license plate. A friend of mine told me that once he actually received a ticket in the mail together with a link. Following the link, he was able to view a video of himself making a right turn without making a complete stop. How amazing!

A BYU parking officer said the following in an interview:

"With the money we saved in parking sticker costs, we were able to buy the car."

What I probably would add to that is: "With the extra parking tickets we were able to write, I am expecting a much bigger bonus!" Just kidding!

Picture of the Day:

Google street view uses facial recognition software agent to detect faces in photos and then blur them for privacy protection. The software agent dutifully blurred hunger striker Bobby Sandss's face in a street portrait in Belfast. (Click the picture to see more!)

My Research: BYU UAV Demo for Utah County Search and Rescue Team

On November 21, 2009, our research group, WiSAR (Wilderness Search and Rescue) demonstrated our UAV technologies to the Utah County Search and Rescue team representatives at Elberta, Utah. Three search and rescue personnel participated in the demo and one of them flew the UAV in a simulated search and rescue exercise.

In two previous blog postings I described BYU research on using UAV to support Wilderness Search and Rescue and UAV capabilities:

My Research: BYU UAV Demo Dry Run
Robot of the Day: UAVs at BYU

The demo was scheduled at 8:30-11:00 am at Elberta, Utah (in the middle of nowhere), which was about an hour's drive from BYU campus. That meant we had to get there by 8 to set up and test equipments. The previous day's weather forecast predicted snow shower, so I was assigned the task of picking up some hot chocolate from the BYU cafeteria so people don't freeze to death!

Despite the facts that I had to deal with my 10-month old son's high fever at 1:30am and not really fall asleep until 3:30am and unconsciously turned off my alarm clock, I actually made it to the cafeteria only 5 minutes late, then I waited another 25 minutes because they haven't made the hot chocolate yet. By the time I arrived at the demo site at 8:30am, turned out the trailer just got there also, so I didn't miss anything! Also, turned out the weather forecast was way off, there was no snow at all, and it was going to be a great day!

Left to right, top to bottom: 1. BYU Cafeteria 2. Beautiful Utah mountains at Dawn
3. The lonely freeway 4. Driving down the highway 5. Good morning, Cows!
6. Gravel road with the destination in view (the ridge in the far distance).

The pictures above were taken by me using an android phone running NASA's GeoCam mobile client. Therefore, all photos were geo-tagged with GPS locations and camera orientation. You can actually view them from Google Earth, where you'll see the exact route I took on the map. Just download the zip file, unzip, and then double click the kml file.

Viewing pictures from Google Earth

The goal of the demo is to show real search and rescue workers how easy and useful our UAV technologies are in support of search and rescue operations. A simulated search and rescue mission was set up, a member of the search and rescue team had to fly the UAV using our interface and locate the simulated missing person (a dummy placed in the wilderness). Students and professors from BYU also acted as aerial video analysts and ground searchers to assist the simulated search. The picture blow shows a ground searcher scouting around in the distance searching for the missing person. The ground searchers always wear bright-colored vests so they can be easily spotted by others (e.g. from the aerial videos) and don't get shot at by hunters. (I know, research is a dangerous profession!)

Ground searcher in a distance (click photo to enlarge)

After setting up everything, Ron Zeeman, a member of the Utah County Search and Rescue team, test flew the UAV, and completed a test drill (launch, manual control, fixed pattern flying, and landing).

Left to right: 1. People busy setting things up 2. UAV at dawn 3. Last minute exercise

After other Search and Rescue team members arrived, we explained how our UAV works, and then started the simulated search and rescue mission. This time I was quite lucky to catch the flying UAV with my camera.

Left to right: 1. Two more professional searchers arrived 2. Two retired UAVs in display 3. The show is on now!

Left to right: 1 and 2. UAV in the air 3. UAV loitering above area of interest (click to enlarge)

Left to right: 1. The kind of junk people would dump to the middle of nowhere 2. Debris of camp fire 3. Real-time video mosaicing (frame stitching)

Eventually, the missing person was located in the aerial video and confirmed by ground searchers. "Unfortunately", by the time we found "him", he was not breathing.

The "missing person" was found, breathless.

Technologies demonstrated include auto launch, auto land, various UAV control mode (carrot and stick, fixed pattern flying, etc.), integrated gimballed camera view in augmented virtuality, click and point gimballed camera control (separate from UAV path), real-time video mosaicing, real-time video annotation and video zooming/scrubbing, point of interest communication between video GUI and UAV control GUI.

Technologies not demonstrated but are work in progress include automatic missing person probability distribution generation, automatic path planning (based on distribution), see-ability metric to measure coverage quality, and automatic anomaly detection.

The demo was a great success! The professional searchers were pleasantly surprised by the ease of operating the UAV and the usefulness of the aerial video support. Their comments included, "That was so cool!" "This could be very helpful!"

Video of the Day:

If the UAV sitting in our lab had a mind of its own,
it would have been singing this all night long...

Paper Review: Distributional Clustering of Words for Text Classification

This paper was written by Baker from Carnegie Mellon and McCallum from Justsystem Pittsburgh Research Center, published at 21st annual international ACM SIGIR conference on Research and development in information retrieval, 1998.

This paper applies Distributional Clustering to document classification. Distributional Clustering can reduce even space by joining words that induce similar probability distributions among the target features that co-occur with the words in questions. Word similarity is measured by the distributions of class labels associated with the words in question.

The three benefits of using word clustering are: useful semantic word clusterings, higher classification accuracy, and smaller classification models.

Clustered Word Clouds by Jeff Clark, perfect in memory of Dr. King on this year's Martin Luther King Day!

The paper first went over the probabilistic framework and Naïve Bayes, and then suggests using weighted average of the individual distributions as the new distribution. Kullback-Leibler divergence, an information-theoretic measure that can be used to measure difference between two probability distributions is introduced, and then the paper uses “KL divergence to the mean”. Instead of compressing two distributions optimally with their own code, the paper uses the code that would be optimal for their mean. Assuming a uniform class prior, choosing the most probable class by naïve Bayes is identical to choosing the class that has the minimal cross entropy with the test document. Therefore, when words are clustered according to this similarity metric, increase in naïve Bayes error is minimized.

The algorithm works as the following: The clusters are initialized with the M words that have the highest mutual information with the class variable. The most similar two clusters are joined, then the next word is added as a singleton cluster to bring the total number of clusters back up to M. This repeats until all words have been put into one of the M clusters.

The paper uses three real-world text corpora: newswire stores, UseNet articles and Web pages. Results show that Distributional Clustering can reduce the feature dimensionality by three orders of magnitude, and lose only 2% accuracy.

Distributional Clustering performs better than feature selection because merging preserves information instead of discarding it. Some features that are infrequent, but useful when they do occur, get removed by the feature selector; feature merging keeps them.




I have a dream that one day when I get old, there will be intelligent robots to take care of people like me, so we can enjoy life freely, happily, and independently.

Joy of Life: Volume 1 Chapter 4

Volume One: The City by the Sea
-- written by Maoni

Chapter 4: Late Night Visitor

“Are you thinking about something?”

The little girl sitting by Fan Xian’s right hand side asked with pouted lips while the two servant girls were busy setting up the dinner table. The little girl was a bit skinny and had somewhat darkish skin. Sitting right next to Fan Xian, whose face was almost as pretty as a girl, the little girl appeared even more pitiful.

Fan Xian reached out and rubbed the yellowish hair on the little girl’s head.

“I am thinking about what kind of tasty food you get to eat every day in the Capital City,” he grinned.

The little girl was the daughter of the Count of Southernland, Fan Xian’s younger sister. They shared the same biological father but had different mothers. Her name was Ruo-Ruo.

She had always been feeble since birth. The Old Madam loved her granddaughter very much, so she sent for her a year ago and kept her in Danzhou for recuperation. A year had already gone by and there still wasn’t much of an improvement. The hair on her head still looked sparse. Born into the family of a government official, she never lacked warm clothing or good food. Her symptoms must have resulted from premature labor, not malnutrition.

Fan Xian found himself quite fond of the little girl. Even though he dealt with her using the attitude of an uncle, treating her as an adorable little kid, playing with her, telling her stories, in everyone else’s eyes, such an attitude became the clear proof of a loving brother-sister relationship.

Nevertheless, due to Fan Xian’s awkward status – a baseborn son, which is very different from a legitimate daughter – the servant girls intentionally avoided mentioning anything regarding the other Count’s Manor in the Capital City.

Since the brother asked, the little girl sincerely started counting with her fingers the tasty things she had enjoyed back in the Capital City. The memory of a three-year-old is surely quite limited, so all she did was to repeat candied haws^[1] and dough figurines^[2].

It was quite late when dinner was over. The setting sun was already half hidden by the other side of the continent, and the dense twilight began to envelop the entire manor.

“Alas! Ruo-Ruo, you are indeed a Weak-Weak^[3]!”

“Brother is teasing again!”

“Alright, alright! What story would you like to hear today?”

“Snow White.”

A big grin suddenly appeared on Fan Xian’s face out of nowhere. Luckily there was no one else around; otherwise one would be shocked to spot such a queer grin only capable by grown-ups appearing on the face of a four year old boy.

“How about a ghost story?”

“No! I don’t want it.” Fan Ruo-Ruo was quite frightened, shaking her head vigorously. Tears quickly formed around her eyes and two streams of tears soon rolled down her darkish face. Evidently, she had been tormented by ghost stories many times in the past year.

Teasing this little girl was only one of Fan Xian’s many vulgar hobbies. What he was best at was bantering with those servant girls. He frequently told ghost stories to those youthful blossoming girls, who always ended up jostling tightly into each other’s embraces screaming and shivering on top of the bed.

Although it was out of the question for Fan Xian to flirt with the young girls vocally for the sake of concealing his true self, he always enjoyed the sweet and tender hugs at these times.

He would always reassure himself with the argument that as a young kid, he was still in a phase where touching was very desirable. So what he did was completely normal and warranted, not something shameful.

Every time the servant girls became curious as to how he could have known so many terrifying stories, Fan Xian would always make his teacher liable. The direct result was that all the servant girls now eyed the teacher with a disgruntled look. “The Count is paying a handsome salary for you to give lessons to the Young Master, yet you teach him ghost stories. It is already evil to scare a little kid, but it is even more evil to scare us, the blossoming flowers.”

After the usual evening ghost story telling was over, the two servant girls, still carrying a mixture of frightened yet satisfied looks, attended to the little guy’s evening hygiene routines and then shut the door to let him sleep.

It seemed to be just another ordinary night.

Fan Xian pushed the hard and uncomfortable porcelain headrest^[4] to the side, and then took out a winter robe from the chest of drawers. Folding it into a nice rectangular shape, he made himself a “pillow”.

He rested his head against the “pillow”, but his two eyes remained open. They shined dimly in the dark night as he remained awake for a long time.

He had accepted the fact that he was reborn into this world, but he was still not used to the customs. It should have been around nine o’clock in the evening, and it was not a comfortable feeling to sleep so early. Besides, he had already slept too much on the sickbed in his previous life.

He stroked the surface of the bed board with his hand and felt better with the conclusion that no one could easily spot the secret casing he had made. As he became more relaxed, naturally, the inner energy inside him began to circulate gradually and he slowly approached the state of meditative trance.

“What kind of life should I live in this world? And how should I spend the next few decades? Maybe I’ll even have many wives and concubines like an ordinary noble man,” Fan Xian’s mind wandered.

Just on the edge of entering the emptiness of mind, Fan Xian was suddenly awoken by an unannounced visitor.

……

……

“Are you Fan Xian?”

A man appeared by his bed all of a sudden. With a grain of abnormal brown in the pupils, his eyes only showed the coldest apathy, clearly indicating his indifference to life.

The question was actually asked nicely. But if such a question was asked by someone who had sneaked into your bedroom at midnight, wearing a mask on his face, holding a dagger in his hand, with a few small bags pinned to his waist, the question could be very terrifying.

If Fan Xian had been a real four-year-old boy, he would surely have screamed on the top of his lungs at the first sight of this queer uncle.

Even if he did all his thinking with his toes Fan Xian could still comprehend that a night traveler capable of sneaking into the Count’s Manor without triggering any alarms had to be someone with high caliber Kung Fu and very possibly, a cruel mind. If he had made any attempt to scream, most likely the man would not hesitate to break his neck in a split second.

At that thought, Fan Xian couldn’t help but feel quite pleased with his calmness. Working hard to suppress the growing uneasiness, he cleared his throat gently. Then putting on the most lovely baby face he could manage, he threw himself forward!

……

……

“Daddy, you are finally home!”

The four-year-old boy dove into the assassin’s arms, tears streaming, and held onto his waist tightly. But the boy’s arms were too short and could not completely circle around the waist. So instead, he grabbed the man’s robe tightly as though he was afraid the man would just suddenly disappear.

Perhaps the boy used too much strength when grabbing the robe, with a tearing sound, a strip of the robe came off.

The man frowned. Without any obvious movement, he suddenly freed himself from Fan Xian’s hug, and then just stood there, dumbstruck, as if he was still trying to work on why this baseborn son of the Count of Southernland had called him Daddy.

Meanwhile, he was also very puzzled. His robe was the top of the line gear of the Bureau. Even a knife wouldn’t have easily cut through it. How was the young kid able to easily tear it apart?

Fan Xian was even more baffled, so much so that he felt as though his heart began to bleed – during times when he managed to be alone, he had always experimented with the power of the nameless inner energy inside him using rocks in the rockwork hill in the courtyard. When he found out that he could almost manage to crush those not-too-hard rocks into bits and pieces with his tender little fingers, he started building some confidence in his self-defense ability.

That wasn’t an easy task to lower the man’s guard with the tears of a four-year-old. Directing all the inner energy to his fingers, Fan Xian thought he had a pretty good chance to subdue his opponent. But who could have imaged that all he was able to achieve was to tear a few strips off the man’s robe.

Something big was bound to happen now.

---

[1] http://www.lannyland.com/jol/TangHuLu.jpg

[2] http://www.lannyland.com/jol/MianRenEr.jpg

[3] The Chinese character for “weak” has the same pronunciation as the character in the girl’s name.

[4] Typical in ancient China before pillow was invented.

Now support the author Maoni by clicking this link, and support the translator Lanny by following my blog! :)

Video of the Day: Matrix running on Windows XP

Click on the picture to view video on YouTube because video embedding was disabled.

Paper Review: A Comparative Study on Feature Selection in Text Categorization

This paper is written by Yiming Yang from Carnnegie Mellon University and Jan O. Pedersen from Verity, Inc. presented at ICML 97 (International Conference on Machine Learning). It has been cited 2742 times according to Google Scholar, another seminal paper indeed!

This paper evaluates and compares five feature selection methods: document frequency (DF), information gain (IG), mutual information (MI), a χ²-test (CHI), and term strength (TS).

A major difficulty of text categorization is the high dimensionality of the feature space. Automatic feature selection methods include the removal of non-informative terms according to corpus statistics, and the construction of new features which combine lower level features into higher-level orthogonal dimensions.

Document frequency is the number of documents in which a term occurs. Terms with low document frequency (less than a threshold) were removed from the feature space with the assumption that rare terms are either non-informative for category prediction or not influential in global performance. Information gain measures the number of bits of information obtained for category prediction by knowing the presence or absence of a term in a document. Terms whose information gain was less than threshold were removed from the feature space. Mutual information becomes zero if the term t and category c are independent. A weakness of mutual information is that the score is strongly influenced by the marginal probabilities of terms. The χ² statistic measures the lack of independence between t and c, and it is known not to be reliable for low-frequency terms. Term strength method estimates term importance based on how commonly a term is likely to appear in “closely-related” documents and is based on document clustering. Average number of related documents per document is used in threshold tuning.

The paper used kNN and LLSF because: both are top-performing, state-of-the-art classifier, both scale to large classification problems, both are a m-ary classifier providing a global ranking of categories given a document, both are context sensitive, and the two classifiers differ statistically. The Reuters-22173 collection and the OHSUMED collection were used for the experiments and recall and precision are used as performance measures.

Experiments include using full vocabulary to removing 98% of the unique terms. Results show that IG and CHI are most effective in aggressive term removal. DF is comparable to IG and CHI with up to 90% term removal. TS is comparable with up to 50-60% term removal. MI has inferior performance due to a bias favoring rare terms and a strong sensitivity to probability estimation errors. Results also show strong correlation among DF, IG and CHI scores.

Video of the Day: Industrorious Clock

Credit: Yugo Nakamura