Hao Wooi Lim's blog

Beauty is (usually) fragile

2013-09-19T03:20:00.003+08:00

While updating my father's iPhone 5 to the latest iOS 7, I can't help but pause and admire the beauty of the phone itself, all the while waiting for the iOS 7 to download. The beautiful, aluminium chamfered edge is just a sight to behold. The phone is light, yet feels solid in your hands. The phone is thin, yet is packed with power. It's unbelievable.

But this is no Apple advertisement. I'm here to tell you that, simply speaking, it all doesn't matter.

What?

Let me explain.

You see, having not own any Apple products now, you might think that I'm unqualified to make this conclusion. But I do. You see, I used to have an MacBook Pro Retina. Until I can't stand it and sold it. My brother had an iPad 2, and my father had Mac mini and iPhone 5. And you know what's my biggest gripe with Apple products so far? It's not that they are not pretty and/or refined. They are. Very much so. The problem is, every time you used it, you just feel like you wanna take a moment and admire its beauty. And I think there are something fundamentally wrong with that. You don't buy a phone to admire its beauty. You buy it because you want to used it to call/sms or whatever. It is just a means to an end. End of story.

I remember back when I had the MacBook Pro Retina, it is simply gorgeous. Plus, it works well as a computer. The only major problem I had with it was that, every time when I try to take it out of a bag to use it, I'm so worried about scratching it. As you know, aluminium is soft and can dent easily. And if you are not careful, you might accidentally scratch it or scuff it. This is just ridiculous. For years using a laptop, I never have that problem. Accidentally dented your laptop? Who cares? It cost maybe RM3-4k and you are gonna buy a new laptop 3 years later anyway. And thus, in the end, people buy cases for their MacBooks and iPhones. I have simply never heard of people buying a case for their laptop. Period.

And thus here lies the fundamental problem with most Apple products. They are just too beautifully made, too fragile to withstand everyday abuse, and too expensive to just get a new one, or to repair whatever damage you have done to it. And in the end, I think it's just sad that people buys a phone as beautiful phone as the iPhone 5, admire it for a few seconds, and then pop that into a case thereafter - hiding its beauty because you feel you had the obligation to preserve that beauty.

It's a tool. Not an art piece. And it should stay that way. Perhaps.

Table of results for NORB dataset

2013-03-21T02:48:00.001+08:00

This is a table documenting some of the best results some paper obtained in NORB dataset.

An Analysis of Single-Layer Networks in Unsupervised Feature Learning (AISTATS 2012)
Cited 161 times. 97%
Regularization of Neural Networks using DropConnect (ICML 2013)
Cited 0 times. 96.77%
Link to project page (Contains source code etc)
Link to Supplementary Material
Link to slides

Table of results for STL-10 dataset

2013-02-19T03:47:00.002+08:00

This is a table documenting some of the best results some paper obtained in STL-10 dataset.

Pooling-Invariant Image Feature Learning (ARXIV 2012)
Cited 0 times. 58.28%
Additional info: 1600 codes, learnt using 2x PDL
Deep Learning of Invariant Features via Simulated Fixations in Video (2012)
Cited 0 times. 56.5%Additional info: Trained also with video (unrelated to STL-10) obtained 61%

Caltech-101 Table of results is updated.

2012-11-03T10:08:00.001+08:00

It has been more than 3 years since I last posted up a table of results for Caltech-101 dataset. I have been updated regularly ever since. But today, it has been updated in full. All the citation counts and fixed a couple of mistakes.

Also, we have a new entry. A paper titled "Spatially Local Coding for Object Recognition" which obtained 81% accuracy, making it near the top of the list.

The irrational hate

2012-09-14T00:54:00.001+08:00

Granted that people likes to hate. Sometimes it is justifiable. Often times, it is not. Take for example this post:

http://www.wpcentral.com/apples-new-nano-nokia-lumia

The claim was that, the new Apple's iPod Nano sometimes has stolen some design cues from Nokia Lumia.

The new iPod Nano looked like this:

Nokia Lumia 800 looked like this:

Wow. It seems to be true then. Apple is copying Nokia. Until you looked further. Nokia Lumia's design actually came from Nokia N9, which had a similar design. The Nokia N9 looked like this:

Nokia N9, on the other hand was released on September 2011.

This is how the previous generation of Apple's iPod Nano (which was released on September 2010) looked like:

Eh, this is difficult, I wonder who is copying who?

Now, I would hate to be engaged in anymore hate fest. I love my android, and I don't think I will be buying any Apple products anytime soon, but this irrational hate fest makes me wanna just go out and buy some Apple products, just so I can screw with those egoistic Apple hater.

Steve Jobs's opinion of Microsoft and my take on it

2012-07-12T22:31:00.000+08:00

In watching "Steve Jobs - The lost interview" (an interview with Steve Jobs made in the year 1995), there was a moment where Steve Jobs talk about what he thinks of Microsoft. Here's a quote:

"Microsoft's orbit was made possible by a Saturn 5 booster called IBM. And I know Bill would get upset with me for saying this, but of course it was true. And much to Bill and Microsoft's credit they used that fantastic opportunity to create more opportunities for themselves. Most people don't remember but until 1984 with the Mac, Microsoft was not in the application business. It was dominated by Lotus. And Microsoft took a big gamble, to write for the Mac. And they came out with applications that were terrible. But they kept at it and make them better. And eventually, they dominated the Macintosh application market. And then used a spring board of Windows to get into the PC market with the same applications. And now they dominated the application business in the PC space too. So they have 2 characteristic. I think they are very strong opportunists. And I don't meant that in a bad way. And 2, they are like the japanese. They just keep on coming. And now, they were able to do that because of the revenue stream from the IBM deal. But nonetheless they made the most of it and I gave them a lot of credit for that. The only problem with Microsoft is they just have no taste. They have absolutely no taste, and what that means is... I don't mean that in a small way, I meant that in a big way in the sense that they don't think of original ideas. And they don't bring much culture into their products. And you say why is it important, well, proportionally spaced font came from typesetting and beautiful books. That's where one gets the idea. If it weren' for the Mac they would never have that in their products. And, so I guess, I'm saddened not by Microsoft's success. I have no problem with their success. They've earned their success, for the most part. I have the problem with the fact that they just make really third-rate products. Their products have no spirit to them. Their products have no... sort of spirit of enlightenment about them. They are very pedestrian. And the sad part is most customers don't have a lot of that spirit either. But they way we are going to ratch it up... our species, is to take the best, and to spread it to everybody so that everyone grows up with better things, and start to understand the subtleties of these better things. And Microsoft is just.. McDonalds. And that's what saddened me. Not that they have won, but that Microsoft products don't displayed more insight and more creativity."

Basically, to sum it up, in Steve Jobs's opinion, Microsoft was a very opportunistic company. They almost always able to identify new emerging markets and they made lousy products but at the same time, Microsoft has this strong sense of self-preservation and perseverance that they will listen to the customer and they will improve their products as time goes (albeit in an incremental sort of way, mostly). That being said, Microsoft products are boring, uninspiring, unradical but nonetheless mostly practical and would certainly get the job done.

And, truth be told, fast forward to present time, Microsoft doesn't seems to be of any different, for the most part anyway. Microsoft products still wasn't cool. Windows 7 certainly wasn't cool. And yes most of us are using it to do pretty useful things. And yes most of us do gaming on the Windows platform. And although Mac OS was indeed gaining ground at the expense of Windows, Windows was still basically the de facto OS. It is which things that we take for granted. For most things, Microsoft was basically a follower, but no one would deny that Microsoft could take the existing things and make it better. One thing that has always held back Microsoft was the unwillingness to break rules. Microsoft was afraid to do something truly radical, something truly different, truly new. Microsoft, in a way they are very much like Google. It is a company run by none other than engineers and marketing people. This is in contrast with Apple, in which designer, thinker and people with passion and vision for the future was dictating the company's future.

The real question is that, should Microsoft be more like Apple? My answer for that would be - no. Microsoft doesn't have to be more like Apple. What Microsoft really have to do though, is to do better at what they are already good at. Microsoft should move quicker, listen to their customer more, make software more secure, cheaper and more user-friendly to the users. And that, would be the end.

Why I would buy an iPhone 4S (Or any Apple products)

2011-10-15T21:23:00.004+08:00

As a gadget lover, I started using smartphone since the days of Windows Mobile with a Sony Ericcson XPERIA X1. It was certainly a very capable smartphone and it is largely usable (well, of course it does). Needless to say, I was utterly blown away when I first gave Android a try. The fact that you can do away with desktop sync, the way you can do away with worrying about needing to copy your old contacts information from old phone to new phone manually is simply magnificent. At that time, I thought no other phones came close, not even the mighty iPhone.

However, one might ask, why is Android the way it is? People who follows the history of iPhone and Android will tell you that they are both introduced on the same year (2007). Well, to be precise, iPhone is first introduced on January 2007, while Android is announced on the same year somewhere in November. If you wanted to know how Android look at that time (2007), this link would shed some light. Now, you would say it doesn't look much like anything like the Android we are used to nowadays. But that was how Android look at that time.

Naturally, the question of whether Android copied iOS or the reverse is debatable. The reality is that both copies stuff from one another. An example of iPhone (as in iOS) copies Android would be the notification center in iOS 5 (the favorite example by Android loyalist). However somewhere in the year 2008/9, Android makes a radical change, abandoning most of the UI before that and fully embrace touch screen, well, just like iOS. By now, I guess it is fair to say that Android borrows quite a lot of concept from iOS. Without iOS, Android *might* have looked different from what it is today.

That is not to say that Android is just like iOS. In fact, Android quite differ from iOS in many ways. For one, the Android provides more flexibility (thus, more fragmented), which many argued that iOS is basically a walled garden system (for better or worse). Apple did this to control the quality of the apps and prevent vendor/telcos from fragmenting the platform by implementing their own UI spin (think HTC Sense UI, etc). Without going into the whole Android vs iOS debate, I guess most readers would recognize their differences and came to the conclusion that comparing them is like comparing Apples and Oranges (Even though both of them are basically fruits and they are very good for you).

However, I have a confession to made. I have respect for people who chooses Android. After all, I did it myself by going with Samsung Galaxy S. The problem is that there is a faction of die-hard Android loyalist that chooses Android because it made them look geekier/nerdier or the fact that it might make them appear smarter than the rest of iPhone-rs (which is most of the people you saw on the streets). While it might be true that they are sort of smarter (or financially poorer, depending on how you look), I believe there is a faction of people who chooses Apple because they appreciate it for what it is. It may be that they are a designer themselves and therefore have better taste, or they knows how to appreciate good UI/UX, or it may be that they just wanted a really good phone. This is something Android loyalist just can't understand. They think that it only attributes to Steve Jobs's reality distortion field or the fact that he's a really good sales man (Even though, that is undoubtedly true).

Personally, as a designer/coder, I have always thought that I had some good design sense. I can appreciate good UI/UX (User eXperience) when I see one. And I strive to design good UI/UX as much as I can. From my experience with using Android and iOS, iOS has slightly better overall UX than Android. Even though both are better than Windows Mobile (I reserved my judgement on Windows Phone 7 until I have used it myself). As such, I certainly can envision myself using both iPhone and Android - most people nowadays carry 2 phones anyway.

In conclusion, there will always be people buying an Apple products just because they wanted to look cool or they simply wanted to brag about it (read: people on Facebook who tells everyone about his cool new phone). However, I have no more respect for people who wanted an Android to look smarter. In the end, the smartest people buys and use whichever phones/gadgets that does what they wanted. There is no need to compare whose balls are bigger - they are all balls.

TL/DR: It's just a fucking phone. Get over it.

Table of results for COIL-100 dataset

2011-08-14T23:45:00.002+08:00

This is a table documenting some of the best results some paper obtained in COIL-100 dataset.

COIL-100 is a dataset containing images of 100 objects, each with 72 views. Here's a link to the official web site for COIL-100.

Nearest Prime Simplicial Complex for Object Recognition (2011)
Cited 0 times. 97.19%
Multiple-View Object Recognition in Smart Camera Networks (2011)
Cited 0 times. 95%
Deep Learning from Temporal Coherence in Video (2009)
Cited 30 times. 92.5%
Deep Learning of Invariant Features via Simulated Fixations in Video (2012)
Cited 0 times. 82%
Additional info: Trained also with van Hataren videos (unrelated to COIL-100) obtained 87%

The Internet is its biggest enemy

2011-07-15T19:16:00.011+08:00

We lived in a new, unprecedented era. An era where Internet is considered by some the most powerful weapon against an oppressive regime. Like Wael Ghonim said, "If you want to free a society, just give them Internet access". If you wanted examples of this, you need only to look at what happened in Egypt, Libya and Syria. But why, you ask. After all, the Internet is merely a communication tool. We've had communication tools for centuries. True. However, you have to remember that communication in the olden days are slow and unreliable. Today we communicate almost at the speed of light and while traditional media like newspaper, magazines and news program can be censored, on the Internet, the more you try to censor something, the more it could backfire, provided the situation is right. This phenomenon is otherwise known as the "streisand effect".

Today, many scandals are being exposed because of Internet. And as companies like Sony, Apple and Microsoft knows all too well, once a bad story turns up on the Internet, there is no going back. You can't take something down on the Internet. Took down one web page and a thousand sprung up. Once an image is tainted, it is really difficult to repair. Of course, this is mostly good for consumers.

In this day and age, it is getting harder to keep secrets. Especially so for secrets that has a profound impact for citizens of a country or consumers. The rise of whistleblower sites like wikileaks would not have been possible without the Internet and an anonymous communication tool such as Tor that allows you to surf the Internet anonymously. Additionally tool like Tor disable an oppressive regime's ability to censor anything on the Internet, not unless the Internet is shut down completely. Even so, the U.S. is working on what they call "Internet in a box" as a way to counter that.

Hence, by now, you must be tempted to think that this is a great era. A great era for freedom. A great era for society that wants to be just and fair for all. However, as great and wonderful the Internet is, it turns out to be its own biggest enemy. Because of availability of anonymity on the Internet, the Internet is filled with rumors, lies and propaganda propagated by irresponsible parties. It is becoming more and more difficult to believe something you read on the Internet and harder still to separate the truths from the half-truths. It is for this reason that most people (older generation perhaps) these days tend to believe what they read on newspaper and/or TV rather than something being forwarded by E-mail, for example.

To give an example, if a particular news that negatively portrays the leader of a particular country has surfaced on the Internet, even if it's the truth, could be rendered (mostly) useless if the people was made to believe that the it was artificially made up by the opposition with ill intentions. The situation is made more futile if no one claims responsibility for spreading the news in the first place. When there is anonymity, there would be no accountability. And the truth is there are people out there that would do all sorts of crazy/illegal things if total anonymity is guaranteed - that is, he/she would not be able to be held accountable, for whatever he/she did. However, in this case, anonymity is the reason the news surfaced at all, because if you know something that could potentially threaten some very powerful people, it is very likely that you would preferred to be anonymous for obvious reasons.

Henceforth we are faced with a difficult situation: we can keep the Internet anonymous and lose accountability, or destroy anonymity of the Internet and lose freedom on the Internet. It would seems that we just can't have it all.

Table of results for CIFAR-10 dataset

2011-02-27T01:29:00.016+08:00

This is a table documenting some of the best results some paper obtained in CIFAR-10 dataset.

Spatially-sparse convolutional neural networks (ARXIV 2014)
Cited 12 times. 93.72%
Additional info: DeepCNiN(5,300) With data augmentation and Network-in-Network layers
Deep Residual Learning for Image Recognition (ARXIV 2015)
Cited 1 time. 93.57%
Additional info: ResNet 110 layers, 1.7 million parameters
Link to slides
Deeply supervised nets (ARXIV 2014)
Cited 66 times. 91.78%
Additional info: With data augmentation. Without data augmentation, it's 90.31%
Link to paper's project page
Network In Network (ARXIV 2013)
Cited 4 times. 91.19%
Additional info: NIN + Dropout + Data Augmentation. Without data augmentation, it's 89.59%
Link to source code at github
Regularization of Neural Networks using DropConnect (ICML 2013)
Cited 0 times. 90.68%
Additional info: Voting with 12 DropConnect networks. With data augmentation.
Link to project page (Contains source code etc)
Link to Supplementary Material
Link to slides
Maxout networks (ICML 2013)
Cited 90 times. 90.62%
Additional info: Consists of 3 convolution maxout layers & a fully connected softmax layer, training data is augmented with translations & horizontal reﬂections.

Link to project page (source code included)

Multi-Column Deep Neural Networks for Image Classification (CVPR 2012)
Cited 170 times. 88.79%
Link to technical Report
Link to Supplemental material
Deep Learning using Linear Support Vector Machines (ARXIV 2013)
Cited 2 times. 88.1%
Practical Bayesian Optimization of Machine Learning Algorithms (NIPS 2012)
Cited 121 times. 85.02%
Additional info: With data augmented with horizontal reﬂections and translations, 90.5% accuracy on test set is achieved.
Least Squares Revisited: Scalable Approaches for Multi-class Prediction (ARXIV 2013)
Cited 0 times. 85%
Stochastic Pooling for Regularization of Deep Convolutional Neural Networks (ICLR 2013)
Cited 34 times. 84.87%
Additional info: Stochastic-100 Pooling
Link to paper's project page
Improving neural networks by preventing co-adaptation of feature detectors (2012)
Cited 261 times. 84.4%
Understanding Deep Architectures using Recursive Convolutional Network (ARXIV 2013)
Cited 4 times. 84%
Discriminative Learning of Sum-Product Networks (NIPS 2012)
Cited 33 times. 83.96%
Beyond Spatial Pyramids: Receptive Field Learning for Pooled Image Features (2012)
Cited 95 times. 83.11%
Learning Invariant Representations with Local Transformations (2012)
Cited 12 times. 82.2%
Additional info: TIOMP-1/T (combined, K= 4,000)
Learning Feature Representations with K-means (NNTOT 2012)
Cited 35 times. 82%
Selecting Receptive Fields in Deep Networks (NIPS 2011)
Cited 51 times. 82%
The Importance of Encoding Versus Training with Sparse Coding and Vector Quantization (ICML 2011)
Cited 202 times. 81.5%
Source code: Adam Coates's web page
High-Performance Neural Networks for Visual Object Classification (2011)
Cited 26 times. 80.49%
Object Recognition with Hierarchical Kernel Descriptors (CVPR 2011)
Cited 55 times. 80%
Source code: Project web page
An Analysis of Single-Layer Networks in Unsupervised Feature Learning (NIPS Workshop 2010)
Cited 296 times. 79.6%
Additional info: K-means (Triangle, 4000 features)
Homepage: Link
Making a Science of Model Search (2012)
Cited 65 times. 79.1%
Convolutional Deep Belief Networks on CIFAR-10 (2010)
Cited 34 times. 78.9%
Additional info: 2 layers
Spike-and-Slab Sparse Coding for Unsupervised Feature Discovery (2012)
Cited 11 times. 78.8%
Pooling-Invariant Image Feature Learning (ARXIV 2012)
Cited 1 times. 78.71%
Additional info: 1600 codes, learnt using 2x PDL
Semiparametric Latent Variable Models for Guided Representation (2011)
Cited 0 times. 77.9%
Learning Separable Filters (2012)
Cited 1 times. 76%
Kernel Descriptors for Visual Recognition (NIPS 2010)
Cited 57 times. 76%
Additional info: KDES-A
Source code: Project web page
Image Descriptor Learning Using Deep Networks (2010)
Cited 0 times. 75.18%
Improved Local Coordinate Coding using Local Tangents (ICML 2010)
Cited 39 times. 74.5%
Additional info: Linear SVM with improved LCC
An Analysis of the Connections Between Layers of Deep Neural Networks (ARXIV 2013)
Cited 0 times. 73.2%
Additional info: 2 layers (K = 2, random RF)
Tiled convolutional neural networks (NIPS 2010)
Cited 33 times. 73.1%
Additional info: Deep Tiled CNNs (s=4, with ﬁnetuning)
Source code: Quoc V. Le's web page
Semiparametric Latent Variable Models for Guided Representation (2011)
Cited 0 times. 72.28%
Additional info: Alpha = 0.01
Modelling Pixel Means and Covariances Using Factorized Third-Order Boltzmann Machines (CVPR 2010)
Cited 84 times. 71%
Additional info: mcRBM-DBN (11025-8192-8192), 3 layers, PCA’d images
On Autoencoders and Score Matching for Energy Based Models (ICML 2011)
Cited 16 times. 65.5%
Factored 3-Way Restricted Boltzmann Machines For Modeling Natural Images (JMLR 2010)
Cited 50 times. 65.3%
Additional info: 4,096 3-Way, 3 layer, ZCA’d images
Fastfood - Approximating Kernel Expansions in Loglinear Time (ICML 2013)
Cited 3 times. 63.1%
Learning invariant features through local space contraction (2011)
Cited 2 times. 52.14%

My predictions for the year 2011

2011-01-05T13:05:00.003+08:00

Well, everyone is posting their predictions for 2011. So for the first time, why not I do one just for the sake of it.

2011 will be the year certain U.S. companies (Facebook?) enters China. Or if not China, maybe some other country in Asia (Japan?). (Sorry, my vision is cloudy for this one)
2011 will be the year Google do social.
2011 will be a bad year for many countries. Many European countries will be entering (or show signs of) economic recession.
2011 will be the year of tablets. Many new models will be launched, as this is the year of Android's new tablet-ready OS Honeycomb. Thus, Android will be gaining ground on the tablet market share, but will not displace iPad as the king of Tablet.
2011 will be the year of the whistle blower. More new alternatives to wikileaks will be launched and make the news, including new shocking leaks.
2011 will be year of peace. There will not be war but more peace talks. There will probably be peace talk: China-US and South Korea-North Korea.

So, to sum it up, keywords to watch out for: China, Social, Tablet, Recession, Wikileaks, Peace.

How to set up OpenCV version 2.2

2010-07-23T02:26:00.007+08:00

If you have been using OpenCV all the time by getting the code straight out of the SVN, you may or may not know that it has reach version 2.2 recently. There has been some big changes to version 2.2 and most of your code is likely won't be able to compile. But, fret not, as I'm going to show you how to get back up again.

First of all, if you used to be doing this:

#include <cv.h>

#include <highgui.h>

just change it to:
#include <opencv2/opencv.hpp>

Secondly, if you used to link against:
cv.lib cxcore.lib highgui.lib
or
cv{version}.lib cxcore{version}.lib or highgui{version}.lib

just change it to:
opencv_calib3d220.lib opencv_contrib220.lib opencv_core220.lib opencv_features2d220.lib opencv_ffmpeg220.lib opencv_flann220.lib opencv_gpu220.lib opencv_haartraining_engine.lib opencv_highgui220.lib opencv_imgproc220.lib opencv_legacy220.lib opencv_ml220.lib opencv_objdetect220.lib opencv_video220.lib
(Some of these libs may not be needed if you did not use the features, but I don't think you will get any linking errors if you put it anyway.)

NOTE: as of the latest in the svn, the version is 2.2. Hence the "220" in "opencv_calib3d220.lib". If the version number changes, you may have the update the number accordingly.

Thirdly, in your "Projects and Solutions" "VC++ Directories" Options diaglog box, add this for "Include files":
{Path to your OpenCV SVN root folder}\include
{Path to your OpenCV SVN root folder}\modules\calib3d\include
{Path to your OpenCV SVN root folder}\modules\contrib\include
{Path to your OpenCV SVN root folder}\modules\core\include
{Path to your OpenCV SVN root folder}\modules\features2d\include
{Path to your OpenCV SVN root folder}\modules\ffmpeg
{Path to your OpenCV SVN root folder}\modules\flann\include
{Path to your OpenCV SVN root folder}\modules\gpu\include
{Path to your OpenCV SVN root folder}\modules\haartraining
{Path to your OpenCV SVN root folder}\modules\highgui\include
{Path to your OpenCV SVN root folder}\modules\imgproc\include
{Path to your OpenCV SVN root folder}\modules\legacy\include
{Path to your OpenCV SVN root folder}\modules\ml\include
{Path to your OpenCV SVN root folder}\modules\objdetect\include
{Path to your OpenCV SVN root folder}\modules\python
{Path to your OpenCV SVN root folder}\modules\traincascade
{Path to your OpenCV SVN root folder}\modules\video\include

Thirdly, in your "Projects and Solutions" "VC++ Directories" Options diaglog box, add this for "Library files":
{Path to your OpenCV SVN root folder}\{The name of the folder you created when building with CMake}\lib\{Release or Debug}

That's all that is to it. Good luck!

Table of results for MNIST dataset

2010-01-17T20:28:00.014+08:00

This is a table documenting some of the best results some paper obtained in MNIST dataset.

Results shown indicates the error obtained by training on all 60,000 samples and testing on 10,000 samples.

Multi-column Deep Neural Networks for Image Classiﬁcation (CVPR 2012)
Cited 9 times. 0.23%
Supplemental material, Technical Report
Deep Big Simple Neural Nets Excel on Handwritten Digit Recognition (2010)
Cited 1 time. 0.35%
Additional info: 6-layer NN 784-2500-2000-1500-1000-500-10 (on GPU) [elastic distortions]
Efﬁcient Learning of Sparse Representations with an Energy-Based Model (2006)
Cited 109 times. 0.39%
Additional info: large conv. net, unsup pretraining [elastic distortions]
Stochastic Pooling for Regularization of Deep Convolutional Neural Networks (2013)
Cited 1 times. 0.47%
Additional info: Stochastic Pooling
Best Practices for Convolutional Neural Networks Applied to Visual Document Analysis (2003)
Cited 190 times. 0.4%
What is the Best Multi-Stage Architecture for Object Recognition? (ICCV 2009)
Cited 39 times. 0.53%
Additional info: large conv. net, unsup pretraining [no distortions]
Deformation Models for Image Recognition (PAMI 2007)
Cited 46 times. 0.54%
Additional info: K-NN with non-linear deformation (IDM) (Preprocessing: shiftable edges)
A trainable feature extractor for handwritten digit recognition (2007)
Cited 38 times. 0.54%
Additional info: Trainable feature extractor + SVMs [affine distortions]
Training Invariant Support Vector Machines (2002)
Cited 281 times. 0.56%
Additional info: Virtual SVM, deg-9 poly, 2-pixel jittered (Preprocessing: deskewing)
Simple Methods for High-Performance Digit Recognition Based on Sparse Coding (TNN 2008)
0.59%
Additional info: unsupervised sparse features + SVM, [no distortions]
Unsupervised learning of invariant feature hierarchies with applications to object recognition (CVPR 2007)
Cited 119 times. 0.62%
Additional info: large conv. net, unsup features [no distortions]
Shape matching and object recognition using shape contexts (PAMI 2002)
Cited 2089 times. 0.63%
Additional info: K-NN, shape context matching (preprocessing: shape context feature extraction)
Beyond Spatial Pyramids: Receptive Field Learning for Pooled Image Features (2012)
Cited 0 times. 0.64%
Convolutional Deep Belief Networks for Scalable Unsupervised Learning of Hierarchical Representations (2009)
0.82%
Large-Margin kNN Classification using a Deep Encoder Network (2009)
0.94%
Deep Boltzmann Machines (2009)
0.95%
CS81: Learning words with Deep Belief Networks (2008)
1.12%
Convolutional Neural Networks (2003)
1.19%
More info: The ConvNN is based on the paper "Best Practices for Convolutional Neural Networks Applied to Visual Document Analysis".
Reducing the dimensionality of data with neural networks (2006)
1.2%
Deep learning via semi-supervised embedding (2008)
1.5%

Table of results for Caltech 256 dataset

2009-10-25T14:19:00.009+08:00

This is a table documenting some of the best results some paper obtained in Caltech-256 dataset.

Results shown here are all trained using 30 samples from each category.

Visualizing and Understanding Convolutional Networks (ARXIV 2013)
Cited 14 times. 70.6% ± 0.2%
Multipath Sparse Coding Using Hierarchical Matching Pursuit (CVPR 2013)
Cited 7 times. 50.7%
Additional info: Multipath Hierarchical Matching Pursuit
Link to paper's project page
Learning Subcategory Relevances for Category Recognition (CVPR 2008)
Cited 48 times. 49.5%
Spatially Local Coding for Object Recognition (ACCV 2010)
Cited 1 time. 46.6% ± 0.2%
Additional info: Multi-scale SIFT features extracted every 4 pixels.
Link to paper's project page
Link to paper's source code
On Feature Combination for Multiclass Object Detection (ICCV 2009)
Cited 376 times. 45.8%
Additional info: LP-β
Link to paper's project page (Contains results, source code and pre-computed features)
Image Classification using Random Forests and Ferns (2007)
Cited 412 times. 45.3%
Local Pyramidal Descriptors for Image Recognition (PAMI 2013)
Cited 1 time. 44.86%
Additional info: P-SIFT + Fisher encoding + SPM + Linear SVM
Link to paper's project page (Contains source code and demo)
A Binary Classification Framework for Two-Stage Multiple Kernel Learning (2012)
Cited 5 times. 44.8%
Efficient Learning of Sparse, Distributed, Convolutional Feature Representations for Object Recognition (ICCV 2011)
Cited 21 times. 42.05%
Additional info: CRBM K=4096
In Defense of Nearest-Neighbor Based Image Classiﬁcation (CVPR 2008)
Cited 478 times. 42%
Additional info: NBNN (5 descriptors)
Locality-constrained Linear Coding for Image Classification (CVPR 2010)
Cited 547 times. 41.19%
Local Naive Bayes Nearest Neighbor for Image Classiﬁcation (2011)
Cited 20 times. 40.1%
Sparse Spatial Coding: A Novel Approach for Efficient and Accurate Object Recognition (ICRA 2012)
Cited 9 times. 37.08% ± 0.36%
Caltech-256 object categoriy dataset (2007)
Cited 596 times. 34.1%
Linear spatial pyramid matching using sparse coding for image classification (CVPR 2009)
Cited 713 times. 34.02%
Kernel codebooks for scene categorization (ECCV 2008)
Cited 242 times. 27.17%

Table of results for Caltech 101 dataset

2009-08-21T22:49:00.088+08:00

This is a table documenting some of the best results some paper obtained in Caltech-101 dataset.

Results shown here are all trained using 30 samples from each category.

Spatial Pyramid Pooling in Deep Convolutional Networks for Visual Recognition (ARXIV 2014)
Cited 191 times. 91.44% ± 0.7%
Link to paper's project page
Visualizing and Understanding Convolutional Networks (ARXIV 2013)
Cited 561 times. 86.5% ± 0.5%
Additional info: Pre-trained with ImageNet
Group-Sensitive Multiple Kernel Learning for Object Categorization (ICCV 2009)
Cited 127 times. 84.3%
Additional Info: GS-MKL
Reference-Based Scheme Combined With K-SVD for Scene Image Categorization (2013)
Cited 2 times. 83%
Multipath Sparse Coding Using Hierarchical Matching Pursuit (CVPR 2013)
Cited 2 times. 82.5% ± 0.5%
Additional Info: Multipath Hierarchical Matching Pursuit
Link to paper's project page
LP-Beta + Geometric blur + PHOW gray/color + Self-Similarity
82.1% ± 0.3%
Learning Subcategory Relevances for Category Recognition (CVPR 2008)
Cited 40 times. 81.9%
Object Recognition as Ranking Holistic Figure-Ground Hypotheses (CVPR 2010)
Cited 68 times. 81.9%
Additional Info: Regression with Post-Processing.
Image Classification using Random Forests and Ferns (ICCV 2007)
Cited 378 times. 81.3%
Additional Info: Bosch Multi-way SVM
Spatially Local Coding for Object Recognition (ACCV 2010)
Cited 0 time. 81% ± 0.2%
Additional Info: Multi-scale SIFT features extracted every 4 pixels.
Link to paper's project page
Link to paper's source code
Link to paper's poster
Local Pyramidal Descriptors for Image Recognition (PAMI 2013)
Cited 1 time. 80.13%
Additional Info: P-SIFT + Fisher encoding + SPM + Linear SVM
Link to paper's project page (Contains source code, demo available)
Sparse Spatial Coding: A Novel Approach for Efficient and Accurate Object Recognition (ICRA 2012)
Cited 7 times. 80.02% ± 0.36%
Additional Info: dictionary size is 4096.
Distance-based Mixture Modeling for Classification via Hypothetical Local Mapping (2013)
Cited 0 times. 80% ± 0.75%
In Defense of Nearest-Neighbor Based Image Classiﬁcation (CVPR 2008)
Cited 442 times. 79.23%
Additional Info: NBNN (5 descriptors)
Smooth Sparse Coding via Marginal Regression for Learning Sparse Representations (2012)
Cited 0 times. 79.11 ± 0.87%
Additional info: dictionary size = 4096
Unsupervised and Supervised Visual Codes with Restricted Boltzmann Machines (ECCV 2012)
Cited 4 times. 78.9% ± 1.1%
Additional info: 1024 codewords trained on macrofeatures with supervised fine-tuning.
Link to paper's poster
Robust Classiﬁcation of Objects, Faces, and Flowers Using Natural Image Statistics (CVPR 2010)
Cited 49 times. 78.5% ± 0.5%
Link to paper's supplemental material
Link to paper's project page
Link to paper's source code (MATLAB)
Visual Geometric Group (VGG)'s implementation of Multiple Kernel Image Classifier trained on dense SIFT, self-similarity, and geometric blur features
78.20% ± 0.4%
Additional Info: Result of 77.8% is obtained by combining dense SIFT, self-similarity, and geometric blur features with the multiple kernel learning
Efficient Learning of Sparse, Distributed, Convolutional Feature Representations for Object Recognition (ICCV 2011)
Cited 19 times. 77.8%
Additional info: CRBM K=4096
On Feature Combination for Multiclass Object Detection (ICCV 2009)
Cited 312 times. 77.8% ± 0.4%
Additional info: LP-β
Link to paper's project page (Contains results, source code and pre-computed features)
Representing shape with a spatial pyramid kernel (CIVR 2007)Cited 547 times. 77.8%
Additional info: Result of 77.8% is obtained by combining all 4 cues (shape 180, shape 360, gray appearance and color appearance.
The devil is in the details - an evaluation of recent feature encoding methods (BMVC 2011)
Cited 91 times. 77.78% ± 0.56%
Additional info: Fisher (FK)
Link to paper's project page (Contains dataset and source code)
Object Recognition with Hierarchical Kernel Descriptors (CVPR 2011)
Cited 29 times. 77.5%
Link to paper's project page (Contains dataset, demos and source code)
Ask the locals: multi-way local pooling for image recognition (ICCV 2011)
Cited 44 times. 77.3% ± 0.6%
Link to paper's supplemental material
A Binary Classification Framework for Two-Stage Multiple Kernel Learning (ICML 2012)
Cited 5 times. 77.2%
Kernel Descriptors for Visual Recognition (NIPS 2010)
Cited 48 times. 76.4% ± 0.7%
Additional info: KDES-A(M)
Local Naive Bayes Nearest Neighbor for Image Classiﬁcation (CVPR 2012)
Cited 17 times. 76% ± 0.9%
Link to paper's associated technical report
Link to paper's source code
Link to paper's project page
Fast approximations to structured sparse coding and applications to object classification (2012)
Cited 2 times. 75.7% ± 1%
Learning mid-level features for recognition (CVPR 2010)
Cited 221 times. 75.7% ± 1.1%
Additional Info: Sparse Codes, Intersection Kernel.
Object and Action Classification with Latent Window Parameters (IJCV 2013)
Cited 0 times. 75.31% ± 0.68%
Beyond Spatial Pyramids: Receptive Field Learning for Pooled Image Features (2012)
Cited 27 times. 75.3% ± 0.7%
Image classiﬁcation with multiple feature (2011)
Cited 3 times. 75% ± 0.8%
In Defense of Soft-assignment Coding (ICCV 2011)
Cited 46 times. 74.2% ± 0.8%
Link to author's web site
Link to paper's source code
Locality-constrained Linear Coding for Image Classification (CVPR 2010)
Cited 446 times. 73.44%
Project web site: Link to Project web site
Source code: Link to MATLAB code (rar)
Linear Spatial Pyramid Matching Using Sparse Coding for Image Classification (CVPR 2009)
Cited 621 times. 73.2% ± 0.54%
Additional Info: Sparse coding, max pooling, linear SVM
Project web site: Link to Project web site
Source code: Link to MATLAB code (rar)
High Dimensional Nonlinear Learning using Local Coordinate Coding (Technical Report 2009)
Cited 3 times. 73.14%
Additional Info: Local coordinate coding, max pooling, linear SVM
Recognition using Regions (CVPR 2009)
Cited 156 times. 73.1%
The importance of Encoding Versus Training with Sparse Coding and Vector Quantization (ICML 2011)
Cited 101 times. 72.6%
Learning Coupled Conditional Random Field for Image Decomposition with Application on Object Categorization (CVPR 2008)
Cited 10 times. 70.38%
Fast Image Search for Learned Metrics (CVPR 2008)
Cited 126 times. 69.6%
Additional info: ML+CORR
A Multi-Scale Learning Framework for Visual Categorization (ACCV 2010)
Cited 2 times. 68.5%
Additional info: sparse coding (K = 900)
Caltech-256 Object Category Dataset (2007)
Cited 544 times. 67.6%
Additional Info: Griffin's SPM
Improved Spatial Pyramid Matching for Image Classification (ACCV 2010)
Cited 3 times. 67.36% ± 0.17%
Variable Sparsity Kernel Learning (JMLR 2011)
Cited 23 times. 67.07%
Deep Learning of Invariant Features via Simulated Fixations in Video (2012)
Cited 2 times. 66%
Additional info: Trained also with video (unrelated to Caltech-101) obtained 74.6%
Similarity-based cross-layered hierarchical representations for object categorization (CVPR 2008)
Cited 47 times. 66.5%
Additional Info: Shapinals
SVM-KNN - Discriminative Nearest Neighbor Classification for Visual Category Recognition (CVPR 2006)
Cited 568 times. 66.2% ± 0.5%
The Hierarchical Beta Process for Convolutional Factor Analysis and Deep Learning (ICML 2011)
Cited 7 times. 65.8% ± 0.6%
Combined Descriptors in Spatial Pyramid Domain for Image Classification (2012)
Cited 0 times. 65.5% ± 0.49%
Bag-of-Features Kernel Eigen Spaces for Classiﬁcation (ICPR 2008)
Cited 2 times. 65.5% ± 0.7%
Image Retrieval and Classification using Local Distance Functions (NIPS 2006)
Cited 136 times. 65.2%
Convolutional Deep Belief Networks for Scalable Unsupervised Learning of Hierarchical Representations (ICML 2009)
Cited 315 times. 65.4% ± 0.5%
Additional Info: CDBN (ﬁrst+second layers)
Beyond Bags of Features: Spatial Pyramid Matching for Recognizing Natural Scene Categories (CVPR 2006)
Cited 2582 times. 64.6% ± 0.8%
Additional Info: L=2, M=200, Pyramid
Source code: Link to MATLAB code (zip)
Slides: Link
Kernel Codebooks for Scene Categorization (ECCV 2008)
Cited 224 times. 64.12%
Visual Word Ambiguity (PAMI 2010)
Cited 257 times. 64.1%
Using dependent regions for object categorization in a generative framework (CVPR 2006)
Cited 121 times. 63%
SIFTing the Relevant from the Irrelevant - Automatically Detecting Objects in Training Images (2009)
Cited 0 times. 61.45%
Pyramid Match Kernels: Discriminative Classification with Sets of Image Features (ICCV 2005)
Cited 849 times. 58.2%
Efﬁcient Classiﬁcation for Additive Kernel SVMs (PAMI 2012)
Cited 11 times. 56.59% ± 0.77%
Max-Margin Additive Classiﬁers for Detection (ICCV 2009)
Cited 79 times. 56.49%
Multiclass Object Recognition with Sparse, Localized Features (CVPR 2006)
Cited 345 times. 56%
Efficiently Matching Sets of Features with Random Histograms (2008)
Cited 46 times. 54.1%
Unsupervised Learning of Invariant Feature Hierarchies with Applications to Object Recognition (2007)
Cited 235 times. 54%
Fast Inference in Sparse Coding Algorithms with Applications to Object Recognition (2008)
Cited 64 times. 53%
Classification using Intersection Kernel Support Vector Machines is Efficient (2008)
Cited 396 times. 52%
Project web site: Link
Source code: Link to MATLAB/C code (tar.gz)
Object Recognition with Features Inspired by Visual Cortex (2006)
Cited 536 times. 42%

Re-focused

2009-04-05T11:54:00.003+08:00

In an attempt to re-focuses my energy on IT-related stuff that I will be doing, I have been doing some thinking and came up with a list (The list is not final):

Research
(I did not say machine learning, nor pattern recognition, as those are just means to an end)
- Computer Vision
- Natural Language Processing

Development
(I did not say functional programming, as those are just means to an end)
- Writing parallel & concurrent programs (Parallel processing, Concurrent programming)
- Writing maintainable, beautiful code

Design
- Designing simple, practical & nice UI

Return to research

2009-03-06T22:49:00.002+08:00

This is a post foreshadowing my come back to the research world. I've been thinking a lot about vision. How do we humans perceive things? How do we recognize things? How do we get a sense of deja vu after seeing things we have seen before?

Even though I do not have the answer to these perplexing questions. I do however think that I *might* have a solution that might solve it to a certain degree. That's a pretty modest statement. But I'm not going to claim anything more exotic before I come up with a working prototype.

The ideas I have in mind have a lot to do with time-based learning, that is, learning not just things but the relationships they have. That is, that certain meaningful things appear to have an order. For example, you do not see a cat walking by, then suddenly it become a dog for half a second, then magically becoming a cat again. Things are not random. They appear in logical order. The idea is nothing new. It has been addressed by Jeff Hawkins in his work on Hierarchical Temporal Memory.

End of part 1. I will talk more about this later.

Programming skill...

2008-12-29T22:46:00.002+08:00

Programming skill...

Those that has it, talks about why there are people who just can't program (no matter what).
Those that don't have it, talks about why the world isn't fair.

Why do I say this? Because if I see another article/blog that says some people just can't program no matter what, I need to pull his head off.

Disclaimer: In no way did I claim I'm superior in programming skills.

Why good is actually bad.

2008-06-08T12:02:00.008+08:00

Basic notion has it that as we gain more experience, our peers or children gets to reap the benefits. Well, the problem is, that is where the problem lies. You see, human learns by experience, be it good or bad. When we made a mistake, some external signals, be it your teacher or your mum tells you not to do it again (Bad experience). When we did something right, you are rewarded or praised (Good experience). Call it reinforcement learning, call it supervised learning, the point is, if you did something wrong, if the signs that you did something wrong was hidden from you, you are learning to repeat your wrongdoing. You may not implied that, or want to, but believe it or not, you are (asking him to repeat the wrong doing). There is no such thing as, "You are doing it wrong, but I'll just let you go this time.". It makes absolutely no sense. If you look at this way, this is common sense. We may not like being scold or make mistakes, it's all aweful medicine, but sometimes the patient just needed it.

The scary thing is, that’s what we do every now and then. Let me tell of a fictional story to illustrate the problem. A software programmer discovered that with C++, it is difficult to write code that does not leak memory, he suggested the team to use Python instead, before the team has decide which language to use. All is good and well. Well, the problem is C++ is still prevalent. Many legacy systems are still written in C/C++. When one of the other programmers in the team was assigned to maintain legacy software written in C++, he might not perform. Why? He/She has never learned best practices of C++! And the sole reason of that is because he/she never gets to learn it anyway; every time the chances come one of the programmer in the team will recommend the team to use a more modern, safer language. So you see, the point is, if you are only told that it's a wrong way and you should not go there, you think that it has saves you whole lot of trouble, but the truly sad truth is, it does not mean that you have actually learn anything. You merely avoided the problem.

Now, some of you might argue that there is simply too much to learn, it is unfeasible to learn everything. While that is true, if that something to learn is something fundamental, it is a mistake not to go through the entire learning process. This is the case where shortcut is not a good idea. Like the famous saying goes, “If we do not learn from history, we are bound to repeat it”. And the only way to learn history is to be in history. Walk in their shoes. Doing what they are doing. Merely being told the mistakes of the past does nothing to prevent it from happening it again, since there may be new, unexplored alternative path that will ultimately lead to the same mistake. And the only way to prevent it from happening is to have greater understanding of the nature of the mistake, and learning how to prevent it from ever happening again.

Perhaps, this is why there has been talk that the quality of Computer Science graduates are dropping, due to the fact that Java was chosen as most CS first language, where as lower level languages like assembly and C/C++ should be taught first, because they not only taught the students about the language itself, but also the fundamentals of computer architecture itself, how things work in the low level and etc. You can read all about it in [Computer Science Education: Where Are the Software Engineers of Tomorrow?].

The truth is, it is extremely dangerous to say that all history is useless experience, though that is not to say that there is no such thing as useless experiences. If you studied machine learning, you will know that, a typical artificial neural network would become brain dead if it is only showed positive samples without any negative samples during the training phase. A human brain works almost in the same principle, since a human brain is really both a discriminative and a generative system. Throwing off the discriminative part of our brain and it is likely that we will never learn. Similarly, if a baby is only taught the right way of doing things, the baby will not have a clue of if what he is doing is right or wrong. In other words, his discriminative ability, his ability to discern what is right or wrong, what is good or bad is simply not there or severely random and thus flawed.

Even if there are times when we needed to choose which knowledge to gain, I shudder to think at the outcome of making the wrong judgement on which knowledge is useless, and which is not: Because depending on what the situation is, it's actually very hard for us to really know for sure. Like Steve Jobs said, it's impossible for us to connect to dots ahead of time. We can only do so, looking back.

This is all obvious and understandable. However, recently, it is increasingly becoming harder to made mistakes, as we are becoming more intolerant to mistakes, while technology is becoming more adept at hiding complexities that are deemed to cause human mistakes. In the office, you are not allowed to use tools that your project leader deemed is bad, you will be fired for making the wrong decisions; In the university, you can't opt for subjects that are deemed to be obsolete by some, you are forced to answer questions based on what is stated on the textbook instead of your own thinking, which ultimately affected our creativity as mentioned in this article; In the house, children are taught not to talk to strangers, just like what Bruce Scheneier in his article "The Kindness of Strangers". In almost every cases, one is simply not acceptable to be different from others, you have to be just like everyone else, even though everyone else could be wrong. We should be given guidance, yes, but more importantly, we should be given the chance to make mistakes! Not repeating the same mistakes should be everyone's responsibility. Not the enforcer. Ideally.

To summarize, discriminative capability is important for us to learn. It is extremely crucial that we are left to made mistakes so that we would not do it again. The experience of doing something wrong is much more important than the experience of doing something correctly. Now, if there is an alternative way for us to experience the whole crucial process of making mistakes without actually wasting any time for it and yet learn a good lesson out of it, I would really like to know about it.

End of examination. Begining of research life.

2008-05-05T23:15:00.002+08:00

Today marks the last day of my examination in this 3 year UTAR life. And I feel like this is the worst examination of all, and the only paper that I have the confidence to get at least A- is AI. To me, doing good in examination just don't motivate me that much anymore, what motivates me more now is research and development in things related to applied AI and computer vision. Which although sounds hard, is actually a lot easier than you might think, because there is no fixed and established way of doing things. Therefore, it tends to tax your creativity more. It's a lot harder to study, understand and memorizes things that other genius scientist/engineer has established, which is what is expected of you in university. In world of research, you don't necessary have to do it using other scientists' ways. They could be wrong and there seems to be always room for improvement. Everyone is solving a small piece of the puzzle, and no one knows the complete solution.

However, before I officially begin my research life, I will take a detour to work in Panasonic R&D for 1 year or so, commencing on June, before coming back to UTAR to begin my Master degree course.

My AI mini project

2008-02-01T10:21:00.000+08:00

UCEC3064 - Intelligent Technique for Engineering Applications

Proposed mini project title:
Improving Circular Pairwise Neural Network Performance on Multiclass Classification Problems

Abstract:
Traditionally, neural networks are trained in a monolithic fashion. That is, a neural network would be trained to classify all K classes, preferably given equivalent amount of training samples for each classes. Training is slow and is unable to take advantage of current multi-cores or multi-processor (SMP) systems to train different classes concurrently. The research done by Teo Choon Hui on circular pairwise classification is an attempt to remedy this by breaking down a K class problem into k binary circular pairwise classification sub-problems. However, this method will reduce recognition accuracy due to the fact that there is a lacked of direct competition between certain pairs of classes. But, at the same time, such a method reduces training time by almost a factor of 3. In this research, we will compare the results of a binary one-versus-all classifier and a pairwise classifier with a circular pairwise classifier as proposed by Teo Choon Hui and attempts to improve recognition accuracy by experimenting with selective circular pairwise classification, in which hard binary sub-problems are paired instead of randomly choosing any 2 classes. We will then attempts to introduce an N-th class into each binary classifier to become a ternary circular pairwise classifier and evaluate its performance.

Keywords:
One-versus-all, Binary classification, Pairwise, Single-winner election methods, Circular Pairwise

PanaGeek final round experience

2008-01-12T20:54:00.000+08:00

It was a rather unlucky day. I got up late, went almost lost searching for PRDCM office. But when I finally found their office, I can't help but noticed the nice view throughout. It's Cyber Jaya after all. When I step into the office (nice office by the way) itself, a bunch of people has already arrived.

When it was time to begin competition, we are given a short tour of the office. Looks good, I saw some huge 24" Dell LCD lying around. We were then proceed to the conference room where we have our round 1 assessment. This is where we are given a scenario and are asked to come out with some ideas for it. I believe I have given a OK idea, although it may not be exactly feasible right now because it involves something as elusive as AI (Actually to be frank, I do not think it's all that hard but somehow I don't think my ideas fare very well in the feasibility department). Due to my deep interests in AI (which happens to be my final year project topic), my idea somehow revolves around it and this competition sort of reflect that.

When it comes to the 2nd round, which is a group assignment, we were randomly group into 4 teams (with 5 in a group) and are assigned a mini project involving some basic socket programming, a typical client/server application. I can say that it's a very simple task if we know socket programming well. Unfortunately, none of us actually do, nothing beyond a last-minute read on socket programming tutorial, that is. The encryption/decryption part could be harder, due to the fact that you need to be able to convert your mathematical skills into programming codes. Which might not be easy if you do not have some background in AI (AI people tends to be good in implementing an algorithm after reading a whitepaper, something that I still have much to learn in this regard). On a side note though, I believe for one to be a good AI researcher, one needs to be an excellent programmer. But that's story for another day :).

At the end of all this, our team got the best team award (the prize was a hamper). And I was awarded a second runner-up, while the other of my team member got a first prize. Congrats Aik Keong! It's a real pleasure working with you. And don't forget my treat after getting your 5k ya, :D.

My programming sucks

2007-12-30T04:08:00.000+08:00

Despite being able to compete in the finals in two separate programming competition (DGX 2001 and PanaGeek 2007), I can't help but thinks that I actually do suck in programming. Allow me to explain.

You see, programming is not just about writing error-free codes. It's also very much about solving the problem using the most efficient way. Computers nowadays may be fast, but there exist problems that could that ages to compute even on the fastest of modern personal computer. Besides that, your code must also look simple. Like they say, the difficulty is not on solving the problem, but solving it and making it look simple at the same time.

To claim that you are a good programmer, you need to be exposed to many different paradigms of programming like functional programming (e.g. Haskell, ...), concurrent programming (e.g. Erlang, ...), logic programming (e.g. Prolog, ...) and etc. For myself, I have limited experience with these kind of languages. I'm currently trying to learn Haskell myself, which is a functional programming language. However, my Haskell journey is far from complete.

If I'm really that good, well, I should be:

Working for Google.
Writing complex 3D games with millions lines of code.
Compete and win in eGenting programming competition.
Compete and win in ACM programming competition.
Solving algorithmic problems in topcoder.com in record time.

Well, all that hasn't happen yet, even though those are some goals I'm trying to achieve. I mean, who wouldn't want all that? All programmers would like to have some of these in their resume, no?

I'm PanaGeek finalist?

2007-12-28T22:05:00.000+08:00

Just got called this morning, informing me that I was selected as one of the top 20 finalists of PanaGeek 2007, along with Lim Fang-Ying and my lab partner Ho Wing Teng.

Glad to know that I was given a chance to compete in the final round. At the same time, I'm also sad. My dear friend, Wing Teng was recently warded into Hospital and just had an operation. His condition is now stable. The bad news is that he may need to stay in the ward for 2 weeks or so. I hope the officials of PanaGeek can made special arrangements for him, or extend the date for the final round. It's important that everyone can compete on equal grounds. I believe no participants would like to win, knowing that he/she won because someone couldn't come that day :).

Panasonic PanaGeek Competition 2007

2007-12-17T08:51:00.000+08:00

I attended the competition on last december 15 2007, Saturday in PJ Hilton. Beautiful hotel and nice food aside, the most interesting thing is the question itself. Here's my take on it.

The questions are slightly more difficult than what I anticipated. Consequently I did not prepared well. Could have used some study on operator precedence and etc. But what the hell...

The questions are broken into 3 sections. The first section aims to test your general C/C++ knowledge. If you can't get pass this, you probably won't do so well in the last 2 sections. In the second section, you have 2 major questions. For the first you get a question on arbitrary multiplication (something I've written before, to solve a n-factorial question). Basically you understand the code, and you answer questions based on it. For the second question, you are shown a flowchart of a hotel room vacant search problem, and you answer a few objective question on it. For the last section, there is 2 problems. For each you are expected to write your solution on the question paper itself. I wish I have slightly more time to complete the shortest path finder problem though *laughs*.

It was a great experience. I enjoyed every moment of it and if I did not win, I'll try again next year.