Truly Converse

Learnings from AfterEamcet Experience

2012-01-01T02:57:00.000-08:00

How it all started? We were inspired by how internet impacted travel agencies in airline industry. With improving access to internet by general public, tickets are being directly sold to passengers. As a consequence airlines don't have to pay commissions to brokers ( or travel agencies). What this means to the passenger is the following:

they don't have to compete with brokers for buying tickets
they don't have to pay an artificially high price ( for the same services)

We imagined a similar analogy with education market. Consultancies and other brokers buy off management quota seats at all the major engineering and medical colleges well in advance. This leaves students and their parents no option but to pay the huge amounts that they demand. To give a more context, here are two situations:

Situation #1:

a medical student with a rank 3000 doesn't get a seat and she can't afford to buy a management quota seat paying half a crore.

Situation #2:

another student who doesn't have a good rank and not even a decent intermediate percentage ( < 60%) gets a medicine seat by paying the huge amount. She goes on to become a doctor to treat us in the following years.

Will the above situations change if we dis-intermediate brokers from the management quota admission process? I bet it would. Here are the reasons:

the prices won't be artificially inflated to such huge amounts; a student in situation#1 might be able to study at a lower price or
colleges might choose deserving candidates as they now have additional information about students. they would have more control over admission process; they might want to conduct an auction. It just won't be unfair; atleast there will be a due-process in-place. Both the parties can take informed decisions. I think it reduces the coercion compared to the previous situation where there is an intermediary.

We wanted all management quota seats to be sold through aftereamcet. We wanted to give students full information about colleges before they choose it. College search was in that direction. There are a bunch of other things we planned - interviews from students, surveys of all colleges, interaction with existing students.

At the same time, we also wanted to get colleges on board. By that time itself, all the management quota for that year have already been sold. They have established processes to handle these seats. We wanted to set a new order of things. We realized that the solution to the problem we are attempting is not in domain of technology; any number of lines of code isn't going to change it. We needed a strategy. Infact, we had well thought ones but we haven't been able to successfully execute them.

We just couldn't take out time and efforts that were needed to tackle the beast at hand. None of us were working full time on this. And in summer, I was working at Google. Similar with other team members as well. In short, none of us had this at the top of priority list.

My learnings form this,

not to point fingers at technology when things don't work; adding a few features won't change anything
to track progress and have numbers on everything; we had an end goal but the path was unclear; we were doing things everyday but we didn't know how much we progressed, whether we are on the right track;
if it's not at the top of the priority list for the whole team, chances are that it will not go much further
to do what I can only do; otherwise it's hard for me to have it at the top of my priority list
to not miss an opportunity to be flamboyant; you wouldn't want to be anything less than flamboyant when you have 30 minutes on stage infront of an international audience ( ISB - ICSEM 2010)

The service is still running and people have been loving map based search we developed: http://aftereamcet.com/index.php/components/collegepredict

My talk at ISB - ICSEM 2010

2012-01-01T01:12:00.000-08:00

At ICSEM-2010 services innovation challenge we finished second runner-up. Also received a cash prize of Rs. 10k. Here is the presentation I gave on that day.

Phone Interview with Incipiens.com

2011-12-31T02:03:00.000-08:00

We applied for funding at incipiens.com. Here are questions they asked.

What are you currently selling? We call ourselves a platform for colleges to conduct their own placement training sessions.

What is your long term vision? How do see your product in future? We would like to open it up for other content providers as well. Training institutes, book authors, individual teachers and the like.

How is your service different from existing learning management systems like moodle etc? We have a superior technology. We developed an adaptive learning platform.

Tell us more about the adaptive learning part? It optimizes two things simultaneously. First is to have no suprises in the tests/interviews, it optimizes learning coverage of student across all topics. It suggest him to explore new topics based on the topics he is already proficient in. Having avoided the surprise he should be proficient enough to answer questions in a topic. To improve his proficiency, our platform continually assesses skill level and suggests him appropriate learning materials; lectures and questions.

What is your business model? Subscription based model. Colleges subscribe for all of their students paying an amount. They can pay in bulk. We already sold to one college;Vignan. We are implementing at various other colleges as well. Or they can collect from the amount from students.

Why will students pay? Most importantly, the training itself is executed by thier own placement officer and not by external vendor. It's endorsed by their placement officer. Secondly, they are already paying huge amounts for campus recruitment training ( Rs. 4000 - Rs. 6000), GATE coaching and other training organized on campus by the college. Thirdly, costs are less compared to existing modes of training. Further, this is reflected in the fact that colleges are willing to buy it themselves - without collecting money from students.

Have you ever paid for any online service while you were a student? We have seen many students pay for IIM-CAT mock tests.

How do ensure quality of content? We are yet to evolve a mechanism for this.

What do need the money for? For growth. ...

Do you have any questions for us? We should have asked questions on a) ventures they already funded b) how they can help us finding US customers and c) what improvements we can make about our presentation.

We are yet hear a reply from incipiens.com.

What advice does Gopi Chand give Saina Nehwal?

2011-12-22T00:14:00.000-08:00

I have always wondered what coaches speak to their wards during the one minute break they get between the sets. Here is a video that captures it; it's from the recent world super series finals between Saina and Wang. Saina lost the finals :(

http://www.youtube.com/watch?feature=player_detailpage&v=e1Vns9ySN7g

See the clip from 52:55 to 53:40 It captures what Gopichand says to Saina between the sets.

It's surprising to me that - at such a level - he says play an 'aaram, simple and alert game'. I have always guessed that inputs would be very much strategic relating to stroke play etc. But here we see an advice concerning more about mind-set than technicalities. And he speaks in Hindi.

ప్రచలిత విధానాలు

2011-12-07T03:30:00.001-08:00

తెలుగు పద అర్ధాలను యాంత్రికంగా గుర్తించడం మీద ఇప్పటివరకు చేసిన ఏ పని గానియు, రచనలు గానియు నాకు కనిపించింది లేదు. అంగ్ల భాషలో దీని మీద గత యాబై   సంవత్చరాలుగా పలు ప్రయత్నాలు జరిగాయి. వాటి సంగ్రహమే ఇ కింది కలము.

ఒక పదం యొక్క అర్ధాన్ని గుర్తించాలంటే ఆ పదం యొక్క అర్ధాలు ఏమిటో తెలిసి ఉండాలి. సాధారణంగా అర్థాల జాబితా నిఘంటువు ల లో దొరుకుతుంది. ఆంగ్ల భాష నిఘంటువులలో ప్రాముఖ్యమైనది వర్డ్ నెట్. ప్రతి అర్ధానికి సంబంధించన నిర్ధారణ, పలు ఉదాహరణలు అందులో లిఖితము.

ఉచ్చారణ ఏ అర్ధ వివరణకి అత్యంత సంభందముగా ఉందొ ఆ అర్దము పద భావాన్ని వ్యక్తికరిస్తుంది. ఎక్కువగా ఏ వివరణ లో పదాలు ఉచ్చారణ లో కూడా ఉంటాయో ఆ వివరణ అత్యంత సంబంధమైనదిగా నిర్ణయించవచ్చు. ఈ తీరు యొక్క అంతర్దృష్టి ఏమిటి?

మనము ఒక పదం యొక్క అర్ధాన్ని దాని చుట్టూ ఉన్న పదాల ద్వార గుర్తిస్తాం. ఉదాహరణకి ఈ కింది వాక్యం లో పంక్తి పర్వత పంక్తి ని సుచిస్తుందో లేక కవితా పంక్తి ని సుచిస్తుందో దాని పక్కన ఉన్న పదాల ద్వారానే తెలుసోకోగలం.

              ఈ   పర్వత పంక్తులు భారత ఉపఖండాన్ని
              ఈ   కవితా పంక్తులు చాలా బాగున్నాయి
              ఈ   పర్వత పంక్తులు భారత ఉపఖండాన

ఇంకో విధంగా చెప్పాలంటే ఒక పదం యొక్క వేరు వేరు అర్థాలు ఉండే ఉచ్చారనలలో ఉండే పదాలు పోలి ఉండవు. ఈ భావన అవలంబించి ఉచ్చారణ ఏ అర్ధాన్ని సుచిస్తుందో తెలుసుకోవచ్చు. కానీ ఈ తీరు కొత్త   పరిదులలో పని చేయదు. ఎందుకంటే అట్టి పరిధులలో

పదాలు కొత్త అర్ధాలు ఉంటాయి.

ఉదాహరణ: ధనలక్ష్మి

పదాల అర్ధాలు కాల క్రమేనా మారుతూ ఉంటాయి.

ఉదాహరణ: భాష

నిఘంటువు లో లేని కొత్త పదాలు కూడా ఉంటాయి. ఈ పడాల అర్ధాలు ఏమిటో, ఎన్ని ఉంటాయో తెలియదు. వాటిని నిఘంటువు లో చేర్చడం చాల సమయం పట్టే పని మరియు క్లిష్టమయినది

ఉదాహరణ: గూగుల్

ఈ సమస్యలను అతిక్రమించేందుకు నిఘంటువులను యాంత్రికంగా సమీకరించడం మొదలు పెట్టారు. ఈ సమీకరణ భాషను బట్టి, పరిధిని బట్టి రుపంతరిస్తుంది. దీనిని అర్థ సమీకరణ అని కూడా అనవచ్చు.

యాంత్రిక తుల్యార్దనం

2011-11-20T09:48:00.001-08:00

యాంత్రిక తుల్యార్దనం

ఒక పదం యొక్క రెండు ఉచ్చారణలు ఒకే అర్దానికి తెలియచేస్తున్నాయో లేదో కంప్యూటర్ కనిపెట్టడమే "యాంత్రిక తుల్యార్దనం". ఇలా కనుగొనడం ద్వార మనం రోజు ఉపయోగించే అనేకానేక సమాచార ఉపకారాలను మెరుగుపరచవచ్చు.

ఉదాహరణకి శోధనా యంత్రాలు ఇచ్చే ఫలితాలలో భిన్నత్వాన్ని పెంచవచ్చు.

శోధనా యంత్రాలు ఇచ్చే ఫలితాలలో భిన్నత్వాన్ని పెంచవచ్చు.

నిఘంటువు లను మెరుగు పరచవచ్చు. వాటి నిర్మాణాన్ని వేగవంత పరచవచ్చు కూడా. తెలుగు లో మొదటి సమగ్రమైన నిఘంటువు ప్రచూరించడానికి నలబై సంవత్సరాలు పట్టింది (1936 - 1978). (http://www.eemaata.com/em/issues/200909/1474.html) అర్ధాలకు ఉదహారణలు సేకరించడం చాల సమయం తీసుకునే పని. అట్టి ఉదాహరణలు యాంత్రికంగా గుర్తించినట్లయితే నిఘంటువు నిర్మన్నని వేగవంత పరచవచ్చు. నిఘంటువు లలో మరిన్ని ఉదహారణలు కూడా చేర్చవచ్చు.

పంక్తి

ఈ కవితా పంక్తులు చాలా బాగున్నాయి
ఈ పర్వత పంక్తులు భారత ఉపఖండాన్ని టిబెట్ పీఠభూమి ని వేరుచేస్తున్నాయి

అర్థ తుల్యనం మనము అలవోకగా చేయగలిగే పని. కానీ యంత్రాలకు మాత్రం అతి కష్టమైనది. మానవ సహాయము ఏ మాత్రమూ అవసరం లేకుండా శబ్ద అర్థాలను తుల్యించే యంత్రాన్ని రూపొందించడానికి నా ఈ ప్రయత్నం

Translating Technical Terms into Telugu

2011-11-19T01:59:00.001-08:00

As I noted in my earlier my post here, I have been writing my thesis in Telugu. Most time consuming task is that of translating technical terms into Telugu. In most cases, I'm translated by associating a new sense to the Telugu word. Perhaps there are no true translations. One needs to find out. The aim of this post is to seek inputs - from all the Telugu speakers out there - to improve the translation precision.

collocate	సంభంధీకాలు
web	అంతర్జాలం
frequency	తరచుదనం
probability	సంభావిత్వం
set	సమితి/సముదాయం
algorithm	పరిష్కారం/రీతి
sense	భావం
meaning	అర్థము
dictionary	నిఘంటువు
induction	సమీకరణ
static	స్థిర
dynamic	వినిమేయ
mutual	పరస్పర
mutual information	పరస్పర దత్తాంశం
machine learning	యాంత్రిక అభ్యాసము
annotate	వ్యాఖ్యానముచేయు
evaluate	నిర్ధారణచేయు
results	ఫలితం
precision	సూక్ష్మత
recall	స్మరణ
stop word/ content words	??
browser	??
web interface	??

Statement of Purpose

2011-11-17T23:01:00.001-08:00

I have been getting a lot of requests to share my statement of purpose. Last year, I applied for a PhD with the below statement-of-purpose. Barring one positive admit result, all my application results were negative. Keeping aside the admit results, I must add that writing this document guided me to develop a micro-philosophy which has been evolving. Another post on it at a later stage! Read and pour in your comments :)

Statement of Purpose

As we read this sentence, we understand with little e ort that the word sentence in its first occurrence refers to a string of words and not to a judgement of conviction. Given an utterance, making computers automatically identify the correct meaning of a word has been the focus of my research work. State-of-the-art systems achieve this by building feature-based representations of each meaning using large number of examples of it. However, for new domains and resource poor languages obtaining such examples is costly. I developed methods specifically to address such scenarios which lack annotated data. Further I employed similar methods - with limited success - to automate the process of building a collocations dictionary. The type of problem solving and problem discovery that I noticed in the course of my research work further enhanced my interest in these areas. Consequently, my plans for a PhD are centered around research in statistical semantics with a special focus on applications arising in digital humanities.

I have been working on Word Sense Disambiguation(WSD) and other problems closely related to it for nearly two years under the guidance of Prof. Rajeev Sangal and Dr. Diana McCarthy. I give a brief account of my contributions towards solving these problems during this time here below

In the recent SemEval task on all words domain specific word sense disambiguation, we harnessed domain knowledge in an unsupervised manner and integrated it in a knowledge based algorithm. Our guiding hypothesis were that 1) domain words are better disambiguators than non-domain words and 2) in general, word senses have skewed distributions and more so in a particular domain. For the environment domain, we extracted key words and estimated their sense distributions in an unsupervised manner. We used this information to bias the random walk on the graph defined over WordNet synsets. Our systems ranked 3rd and 4th in the public evaluation. The top two systems used annotated data from the domain whereas our systems did not use any. Our experiments show that by mining domain corpora one can get results close to that of a supervised system.

My most enriching learning experience has been working on a project addressing the requirements of Macmillan, one of the leading dictionary builders, towards building a collocations dictionary. We set out with an objective to automate the association of collocations with senses. For a given headword, our first step is to extract collocations from the corpus and the second is to assign them to a particular sense of it. The collocations were extracted from a massive web corpus due to which a large number of false positive collocations were seen. To eliminate these, heuristics were defined based on the collocation length and non-existence of collocations in the stop word lists, multiword lists and headword lists. This eliminated most of the false positives. To associate senses with these collocations we utilized the thesaurus that grouped the dictionary senses. With the assumption that a sense shares collocations with its thesaurus members, we performed a biased random walk on the graph whose edges are between all the thesaurus group headwords and their collocations. We achieved a 80% reduction in false positives and an overall 40% precision at 20% recall in associating them with senses. Our preliminary work appeared in the proceedings of Asialex 2009.

Working on this project has been unique in the sense it gave me an exposure as to how research in word sense disambiguation can applied to expedite the process of dictionary building. The interdisciplinary nature of the project drew my attention toward the intersection of statistical semantics and digital humanities. There are many interesting problems at this crossroads that I would like to investigate in my PhD. Specifically, I intend to build tools that assist scholars in doing their work. For example, a literary scholar studying the variations in usage of a word would find a word sense induction system helpful; disambiguating and ranking sentences would assist a lexicographer in choosing right examples for a given sense entry. Multimodal representations of word meaning would enable scholars in understanding documents as more than text. Such tools, by linking textual and archaeological objects, would prompt archaeologists with evidences to form interpretations about the past. This is not to say that humanities is to be looked at merely as an application domain. Humanities by itself, although open-ended, involves serious theorization. I strongly believe the best tools could be built by making humanities an equal knowledge partner in this process. Perhaps we would one day be able to fully simulate scholarly work.

After my PhD, I look forward to taking up a position in academia. In this respect, I'm motivated by my fond teaching experiences for various undergraduate courses. My most enjoyable teaching experience was when I was a teaching assistant for the Introduction to Computer Programming course. I was actively involved in giving recitations, grading exams and creation of test data for automatic grading of assignment problems. Yet, the most satisfying part was in choosing the right examples during the recitations. I was aware that the students had a strong math background so I had consciously chosen examples built on their background. For instance, during the recitation on binary search I chose the example of finding first positive intersection of the curves y = x and y = tan(x). It was highly encouraging to see the students themselves discover the notion of eliminating half the possibilities at each step.

For my consistent involvement and all round contribution both in academics as well as extracurricular
activities I was awarded a gold medal as the Best All Rounder of the Year 2009 of my institute.

Given my all-round background, I believe I am in a good position to make crucial contributions in my pursuits at the intersection of statistical semantics and digital humanities. Interacting with members of your group will give me an invaluable opportunity to broaden my horizons, and work in many research directions that I cannot anticipate at the moment. I look forward to a long and mutually beneficial relationship with X.

Leveraging Technology to Improve CS Education

2011-05-27T10:16:00.000-07:00

To give much needed hands-on exposure to building things, Computer Science unlike other fields doesn’t demand huge infrastructure - all that one needs is a desktop or laptop or a netbook.

While the infrastructural needs of computer science students are minimal it doesn’t in any way alleviate the need for good classroom instruction where the much needed basics are covered. But high quality class room education has always been a scarce resource in our country. Among others, the main reason for this lack in quality is the lack of skilled faculty members. In various institutes the faculty members, themselves, are not equipped enough to prepare students to meet industry expectations; only 10 % of Indian graduates are employable according to NASSCOM reports.

To support the classroom instruction, there have been efforts by NPTEL. These are mainly video lectures and the students are passive listeners to this content. These are not learner centered resources - there is no feedback to student of any kind in this setting. Can we do better than video lectures? Recently, IIIT-H in collaboration with various other institutes started developing virtual labs; these are better than video lectures in the sense students learn-by-doing. There is minimal feedback in this, but this doesn't match a (skilled) human tutor.

Going back to the original question as to how we can give feedback, there are atleast three possibilities to give feedback to student on a course: 1) through human intervention - an instructor gives feedback to student, but this is very costly 2) totally automated feedback - this is hard to realize in all parts of CS and 3) using a mix of both #1 and #2.

What are the ways in which we can realize #3?

My MS Thesis has Zero Pages

2011-05-09T12:19:00.000-07:00

Acknowledgements

As an undergraduate 3rd year student, with all the uncertainty whether I could do research or not, I took up a research project which turned out to a be a total failure. That's the end of the story - I decided not do any more research. What changed my plans was the generous support of one professor, my advisor, Prof. Rajeev Sangal, who treated me like any other research colleague and supported me for another project even after that. My debt to Prof. Rajeev for his support and expert counsel without which I wouldn't have been experiencing the joys research ever since.

Even before I knew I had been working with the author whose papers I have always admired, I co-authored my first paper with the same author, Adam Kilgarriff. My first association with him still remains unforgettable - I responded to his email (on corpora-list) in which he posed a word sense disambiguation task. He accepted my proposal and gave me the project. I was filled with immense gratitude toward him and the community since my ideas are judged rather than my personal background. During the project work, I got to know the nuances of word sense disambiguation which greatly influenced the direction of my research work.

Beyond doubt I would say this work wouldn't have been possible without Siva Reddy, my project mate who is a unique combination of scientific and interpersonal humility. I owe a lot to him for suggesting me to work on this thesis problem and for all the productive discussions we have had in which we re-discovered many ideas in the extant word sense disambiguation literature and for all the dinner table conversations during which he taught me the art of socializing. I couldn't have asked for a better project mate than him.

I'm grateful to Dr. Diana McCarthy and Dr. Mark Stevenson for their expert guidance and constant encouragement. All our discussions were through long e-mail conversations. Several ideas in this thesis have their roots in these discussions.

I owe a great deal of gratitude to many people who have enriched my undergraduate experience: busy as he was, Pranav Kumar Vashista always had time for me to discuss life, universe and everything; Siddharth Ravi for being my advisor since the early days of my stay at IIIT; Poornima and Ravi Kiran for providing a much needed boost when I was feeling low; Himanshu Singh, Kartheek N V and Rahul Katragadda for all discussions and debates with opposing viewpoints during which I was corrected wherever required; Avinesh PVS, Chandrakanth, Kranthi Reddy, Naga Satish, Phani Chaitanya, Prasant Gopal, Prudhvi Vatala, Ravi Raj, Saladi Rahul, Sharat Chandra, Spandana Reddy, Sunil Soni, Venkat Srikanth, Vidyadhar and Vikranth Kumar for their faithful support and instant help even though I was largely forgetting them during my frequent academic seclusions. And for all those who were part of my tutorials. My five years at IIIT, Hyderabad wouldn't have been as much enjoyable without all of them.

Above all, I thank Navjyoti Singh, one of the most amazing person I have ever met, for often challenging me; my sister, Anupriya, who makes me really proud everytime she tells me about the little everyday discovery she makes; and my parents for their unconditional love, support and belief in my abilities.

ps: Coming to the title of the post, I have decided to write my thesis again. If at all I submit, it will be a thesis written in Telugu. From now on I will call it "పరిశోధన".

రంగుల పదం రంగు ఏంటి?

2011-02-24T00:14:00.000-08:00

పదాలకి రంగులుంటయా? ఉండటమే కాదండోయ్, ఉసరవెల్లి వలే పదాలు కూడా వాటి రంగులను మర్చుకుంటాయి అని అంటాను నేను. ఉదాహరనికి కాలు అను పదానికి (కనీసం) ఈ రెండు రంగులు వాడుకలో ఉన్నాయి:

'ఆమె' అసెంబ్లీలో కాలు పెట్టదా? తిరిగే కాలు తిట్టే నోరు ఊరక ఉండవు
దోసె కాలడం పూర్తవుతుండగా ... ఈ విశ్వ మంతటిని కాల్చు ...

మనము వేరెవేరు వాక్యాలలో పదాలను వేరువేరు రంగులలో (అర్థాలలో) ఉపయోగిస్తుంటాం. అయినప్పటికి వాక్యాల్ని చదువుతున్డగానే మనము అతి సులువుగా అందులోని పదాల యొక్క రంగుని (అర్థాన్ని) గుర్తిస్తాం. కానీ కంప్యూటర్లు అట్టి సులువుగా అదే పని చేయలేవు.

ఉసరవెల్లి ఏ విధంగా ఇతే దాని చుట్టూ ఉన్న పరిస్తితుల్ని బట్టి ఏ రంగుకు మారాలో గుర్తిస్తుందో, అటులనే కంపూటర్లు ఒక పదం రంగుని గుర్తించడానికి దాని తోటి పదాల ఏమిటో గమనిస్తాయి. అందుకే ఒక పదం యొక్క రంగుని కచ్చితంగా గుర్తించడానికి వాటికి అనేక ఉదాహరణలు అవసరం. ఇది ఉత్తమమైన తరీకా. కానీ ఉదాహరణలు తయారుచేయడం ఖరీదయిన పని. నిధి దాతల కోసం వేఛి చూడాల్సిందే! ఆంగ్లము లో దినిని సుపర్వైసడ్ వర్డ్ సెన్స్ డిసఅంబిగుషున్ (Supervised Word Sense Disambiguation) అని అంటారు.

కొన్ని సార్లు ఉసరవెల్లిలు వాటి చుట్టూ ఉన్న ఉసరవేల్లిల రంగు ఏంటో చూసి వాటితో సమానంగా ఉండే రంగులోకి మారుతాయి - దినిని మనం సాంఘిక సంకేతం ( సోషల్ సిగ్నల్లింగ్) గా చెప్పుకోవచ్చు. అటులనే కంపూటర్లు ఒక పదం చుట్టూ ఉన్న పదాల రంగుని చూసి దాని రంగు ఏంటో గుర్తిస్తాయి. ఇలా గుర్తించడం ఒక చిక్కు ముడి విప్పిన దానితో సమానం. ఎందుకంటే చుట్టూ ఉన్న పదాల రంగు ఏంటో తెలియాలంటే ఈ పదం యొక్క రంగు ఏంటో తెలియాలి. దినినే నాలేజ్ బెసడ్ వర్డ్ సెన్స్ డిసఅంబిగుషున్ (Knowledge Based Word Sense Disambiguation) అని అంటారు. ఉదాహరణలు అవసరం లేకుండా రందులు గుర్తించడం ఈ తరీకా ప్రత్యేకత.

ఇప్పటి వరకు పై రెండు తరీకాలను కలగలిపిన తరీకా పై పరిశోదనలు జరగలేదు. నా తదుపరి వ్యాసాలలో వాటి మీద నేను చేసిన ప్రక్రియలను వివరిస్తాను.

My Hypothesis on Curing Cramps

2010-10-08T12:01:00.000-07:00

Summary: Muscle cramps, also known as Charlie Horses or spasms, are excruciatingly painful. No one knows the exact reasons as to what causes cramps. And a cure for muscle cramps is still very far off. This is a hypothesis designed to find a cure for cramps. This is inspired from Mirror Box experiment designed by Vilaynur S Ramachandran and from my personal experience in with dealing with cramps of fellow football and badminton players.

What is a cramp? Cramp is a common injury for athletes at all levels and affects hard working muscles such as the calves in runners and the arm in tennis players. But despite being able to cure all kinds of killer diseases, scientists still don’t know what causes it?

Mirror Box? A mirror box is an apparatus to relieve a patient of pain experienced from an amputated limb - even though the limb doesn't exist a subject still experiences the pain. A diagrammatic explanation of the mirror box.

The patient places the good limb into one side of the box (in this case the right hand) and the amputated limb into the other side. Due to the mirror, the patient sees a reflection of his good hand where his or her missing limb should be (indicated in lower contrast). The patient thus receives artificial visual feedback that the ”resurrected” limb is now moving when he or she moves the good hand.

The pain here is caused due to a conflict between visual feedback and proprioceptive feedback. And this conflict is nullified by using this apparatus. These are also called as phantom limbs.

Insight: Both cramps and phantom limbs share commonalities: they are both painful, they usually occur in one hand or one leg and both re-occur. May be the pain incase of cramps is also induced by a conflict between visual feedback and proprioceptive representations of the cramped leg. It has the following interesting implication: the imagery of movement of cramped leg alleviates the pain - but we can't move the cramped leg as it causes even more pain. So what do we do - put a mirror?

The method that I employed can be summarized in one line as follows:

perform an action (look, touch, etc ) on left leg. Right leg gets relaxed automatically

The actions/instructions that I gave to my friends in such cases

I told them, have a look at your left leg it’s perfectly fine
I touched/patted his left calf to give tactile inputs

As already said this is just a hypothesis and NOT a proven cure. A scientific study is needed to validate it.

ps: this was my cognitive science weekly report.

ATM Erroneous Debit

2010-10-07T04:50:00.000-07:00

I was recently at an ATM which hanged for quite sometime as soon as I began to operate it - to withdraw an amount of Rs. X. It didn't even give out a slip to acknowledge my transaction. Till here I thought nothing went wrong. But I was shocked to see (online) that an amount of Rs. X got debited from my account. A loss of Rs. X for me - I hope the bank will prove me wrong.

The first thing I did as soon as I realized that a "erroneous debit" happened in my account is to search in Google "ATM erroneous debit". I found that there were 17 hits - although very less it's an assurance for such cases have been previously reported. Further I got the information on whom to complain to and how much time it would generally take to process such complaints.

Later I went to the bank manager and told him the situation. He says it is hard for him to launch a complaint with out transaction id. He said I did a big mistake by not bringing the debit slips of that withdrawal. How can I bring the transaction slips when the machine itself didn't give me one? He finally agreed to forward my complaint after teaching me the security issues related to customer (mis-)use of ATM.

I wish the banking organizations have a very less fault recovery time.

Two Costly Myths on Engineering Branches

2010-06-23T10:02:00.000-07:00

I intend to address two common myths that students have in mind while choosing a branch:

Some engineering branches are bad: In all the branches you study in order to build and you build in order to solve a real problems. Then if all them are doing the same thing how do they differ? Consider the following metaphor. Imagine a room with multiple windows and with a number of objects int it. Each of these objects would look differently when looked at from different windows - we get to view them in different shapes, colors and sizes. Similarly, engineering branches are like the windows from which the objects called real world problems are analyzed from different perspectives. For example consider the relief operations at Yushu and Haiti earth quakes: (to name a few contributions) chemical engineers made plans for improving access to clean water; food engineers made sure the quality of food remained intact; doctors in medical campus used computers to translate Haiti utterances to English; electrical, mechanical and bio-medical engineers came together to put up solar suitcases to power the hospitals; and earth quake engineers suggested simple and cost-effective ways of rebuilding Haiti. To re-iterate my point all branches train with the skills necessary to solve real problems.
There are easy branches: If we look at the history of each of these engineering branches, all of them have atleast a 50 year history of knowledge discovery. For example, computer science dates back to atleast 1956 historical meeting on Artificial Intelligence; electronics dates back to the telegraph industry of 1800s during which radio and telephone were discovered;aeronautical engineering started developing at rapid pace since world war 1, a decade after Wright brothers flew their first flight (similar tales can be found with other branches as well). Now, for the one who intends to build systems in his/her respective discipline it is necessary to have a thorough grasp on the foundational developments of the past. From the rich history, it seems none of them are easy to study and it is only our interest and deliberate effort that makes them easy.

These two have become costly myths in the recent days. These myths are 1) discouraging bright students to enter the 'lowly' ranked branches and thus hindering development in those areas, 2) conditioning the interested students to abadon their natural interests. This is something on which the course instructors and senior students should give a serious thought - we got to change the public image of the engineering branches.

My Girl Friend

2010-06-18T02:45:00.000-07:00

Disclaimer: The author has no intentions to offend any girl.

I had a wonderful girlfriend. She is so sweet that her thought itself mesmerized me, drove me crazy. The relationship lasted for more than 3 months. Finally, she says, "I already have a boy friend, I don't have patience to maintain two boy friends."

My girlfriend was both dishonest as well as honest, dishonest toward the relationship and honest in conveying what she felt. Ofcourse, I was displeased with the situation, not with her dishonesty per se, but with the unpredictability when it comes to things which are important to me. In retrospect, I realized that this relationship was unpredictable and luring.

There is another relationship that shares commonalities with this one. That is between me and my grades.

Both of them are highly unpredictable: The grades are unpredictable because it intends to measure human capability, which cannot be crisply defined in the first place, in an extremely conservative manner by objectively measuring my proficiency in testable skills through various means, viz: assignments, quizzes, surprise quizzes, mid-term exams, end-sem exams, lab exams. On top of it, all these time constrained tests discouraged me to work at my own pace and leisure.

Both of them are luring: I think it's not uncommon that guys ogle at girls, but, too much of ogling might be injurious. Have (you) ever been slapped by a girl? In the relationship with my grades, I was lured by the attractive baggage of after benefits, ranging from monetary to matrimonial, it brings with it. So, I didn't want to see any scars on my grade sheet. Such motivations are not bad, but I fell in trap only working towards achieving proficiency in testable skills. It took me time to realize that a course is much more than that.

Personal realization made me appreciate things much better. I learnt that in a place like IIIT Hyderabad, which nurtures big ideas the emphasis of a course is not merely on the grades, but it is in making students absorb big ideas and to apply craft solutions to complex problems.

I wish I had absorbed this idea to de-emphasize grades and direct my efforts to absorb the big ideas long back.

ps: I wrote this article nine months ago. I was in a dilemma whether to publish or not. Finally, I did - when you are a student you are excused for drawing such comparisons - compare grades and dishonest gf :P. I hope the matured readers excuse me for this!

Selling Sensitive User Data to Third Parties without Compromising Privacy

2010-06-17T06:35:00.000-07:00

(A post written in haste!)

There are cases where the organization have sensitive private data of users such as phone numbers which they want to share with third parties ( potential advertisers ). However, they don’t want to give away these details directly in its original form. Users on the other hand might agree to this if it is not a compromise on privacy and such a thing should be done after taking prior consent of the user. How then might the third parties communicate with the users (potential customers) without compromising privacy?

Here is an implementable model. The new system works like this :

organization will give two numbers to the third party

organizations number
alias of user X

the third party would first call the organization and then connect to the user using the alias.
Initially, organization will generate aliases corresponding to the user numbers
sells the aliases instead of the original numbers; these aliases expire after the call exceeds certain amount of time.

How can this be implemented? A simple software using which an organization can

buy X Skype numbers. X depends on number of potential advertisers and customers.
generate user phone number aliases [and then sell these to third parties :P]

In the back-end when a third party calls the organization number - it will be simple table look up and then forwarding the call to the respective user using SKYPE-API.

Note: Such Skype numbers are not available in India- but I’m sure there will be some alternative.

ps: I am not really into software market and I have no plans of jumping into it either. Do let me know if you happen to implement this thing or find such an implementation :P

iList: Intelligent Linked List Tutor

2010-06-12T09:37:00.000-07:00

http://www.digitaltutor.net/

I'm thrilled with this! Totally loved it. Any one who has coded linked lists will surely appreciate this tool. Do try it. For first timers this is a fun way of learning linked lists.

For those of you who have familiar with Assignment evaluator ( a.k.a c-programming portal) It is similar to our portal to the extent both of these pose problems; this one goes one step further by giving feedback at code level.

It uses artificial intelligence technology to achieve this. For more details read here.

ps: if you are aware of any similar tools please don't hesitate to add them in comments.

AfterEamcet.com: A students companion for EAMCET counselling

2010-05-30T07:37:00.000-07:00

In Andhra Pradesh, aspiring engineers write EAMCET, Engineering Agricultural and Medical common entrance test for an admit in one of the 753 engineering colleges across the state. Before they get an admit and after they get their rank is the crucial phase of counseling in which the college assignment is made based on the student rank, availability and his choices. As students can't visit all the 753 colleges to decide the one which best suits their individual preferences most students end up filling choice list in a sub optimal manner.

We propose an IT services based solution to this problem. Student can create a personalized list of college choices using our methods. Unique features of our solution are:

EASY: Thousands of students. Hundreds of colleges. One stop site where all the information about colleges is available.
FAST: Narrow down your college options in a blink of an eye. Filter your results based on the branch, distance to the nearest bus-stop, college transport facility, hostel facility and more
PERSONAL: Students can search for colleges nearest to his hometown or location of his/her choice.
FUN: Student can identify their engineering-twins: other students who have similar personal preferences.

The current version can be accessed at aftereamcet.com. This is the only website which has all the above listed features till date.

ps: Encourage students going for EAMCET counselling to visit aftereamcet.com

First Days at Brno, Czech Republic

2010-05-10T02:00:00.000-07:00

It has been three days since we arrived at Brno after a long journey of 28 hours ( Hyderabad -> Bangalore -> Paris -> Prague -> Brno ). Adjusting to local weather conditions and getting to terms with the structures here has been a thrilling experience. Here we find lot of interesting things: TRAMs that stop to let us cross the road at the zebra crossing, buses which look like trains;they run on electricity with two overhead electric wires, swipe cards at all the entrances; we don't see security guards at these, long days from morning 6:00 a.m to evening 8:00 p.m, temperature that is atmost 15 degrees during day, sugar written as 'cukor' in malls, chilli sauce that is atmost as hot as tomato sauce, one has to pay a fine of 500 crown for ticketless travel in TRAMs and buses, tap water that is safe for drinking, KFC and other restaurants offering free cool drinks such as coke, pepsi and the like; in this chilly weather people prefer 'warm' drinks to 'cool' drinks, a few Indians, fewer Indian restaurants and that we have to cook food ourselves.

Four of us (Avinesh, Girish, Siva and me) have never cooked anything. It is first time for all of us in the kitchen. The common kitchen in our dormitory had an electic stove and an oven. We prepared a menu for ourselves, an ambitious one: rice, chicken curry and aloo fry. We went to a nearby store to buy the necessary ingredients and other containers for cooking. After a two-hour long struggle in our kitchen we finally tasted a delicious meal. Chicken curry and Aloo fry are big hits. But we messed up while cooking the rice - the lower layers in the bowl got roasted up in the bowl, the upper ones remained unboiled and the middle ones became a gluey paste. Boiling rice isn't an easy task!

Kudos to our cooking team: Avinesh - head cook, Girish - critique, Siva - resource manager, Abhilash - vegetable cutter. We can also cook :P

Tricks and resources that helped me in doing my research!

2010-04-29T03:47:00.000-07:00

Genesis of my problem statements: There are two strategies I used for choosing research problems in my area.

Attack Open Problems in the Field: I have chosen to work on an AI-hard problem in my area which is unsolved for the past 50 years. And it remains still open and we studied a few aspects of the problem which resulted in a few publications. I came across this problem while I was attending IJCAI-2007 hosted at our institute. Such problems are so hot that conference and journal papers repeatedly attempt to solve them. It might not be advisable to tackle such problems without having some sort of idea of yours. In our case, we used ideas from Multi Agent Systems to solve our problem.
Devising Tractable Avatars of Open Problems: This is another trick I have seen repeatedly used in my field. The motivation behind devising this is as follows: To solve an open problem X, it better to first solve the tractable avatar of it namely Y; Y might lead to solving X at a further stage, say, 10 or 15 years down the line. Y has the remarkable property of being tractable. Further, Y is chosen in such a way that it has many, many useful practical applications. Now the question of solving Y becomes obvious;the solutions attempted to solve X can be easily adapted to solve the simpler problem Y. By no means, I’m saying this adaption is trivial, but it's one strategy that can be used while looking to solve tractable research problems which have practical significance.

The research process:

Most of the work I did is in a collaborative fashion. I do a part of work and the other authors do rest of it. It reduced the amount of effort I had to put in. It might sound as if we did things in a planned manner. We were actually least planned. We did most of the research work in a rush and over-enthusiastic manner. As soon as I have a new idea, I just go and discuss it with my mates. At times, we waited discussing the merits and demerits of the idea. But more often, we just implemented it then and there itself within a few hours before the enthusiasm gets down. We failed often but we learnt how to solve the problems.

The whole process can be summarized as follows:

Discuss the idea with my colleagues ( who can help me )
Implement the idea and check the results within few hours ( I was afraid that I was spending too much time on a non-working hypothesis. So we finished off things as soon as possible. Since it is a short span we never faced problems of low motivation. )
If the results are worth publishing, we did that. Otherwise, we try a new hypothesis or abandoned the problem itself.

Tools and others I found useful:

Aspell: to do spell checking for those who write in Latex on unix machines.
The ”Cited by X” link provided by google scholar at the bottom of every search result.
Latex Beamer/Prosper Package.
Power Point principle: ”To use power poing effectively, you should have no more than five points per slide and no more than five words per point”
Email: Writing is more concrete than audio signal. Most of our ideas were actually discussed through email. At the end of discussion the idea was concrete enough so that it can directly be implement.
MIT Open Course Web (OCW): The high quality materials here are always a resource for me to look for.
Although I’m a big fan of computer, I have reservations to using it while my ideas are in developing stages; it cannot be a substitute to working with a pen and paper.

Experiences I missed in my B.Tech life

2010-04-24T05:00:00.000-07:00

I have long wanted to write this post! One of my favorites. It lists experiences that I have missed during my B.Tech life.

A stage performance on cultural night or felicity
Talking more to girls (Both batchmates and others )
Becoming friends with more professors
Learning to solder ( during ECE labs )
Taking classes for fun (dancing? creative writing? )
Learning an instrument (guitar? )
Start a company ( does it matter if I fail? )
Using library resources
Volunteer with children at Ashakiran.
Having lunch or dinner with my professor
Maintaining my personal webpage
Getting on to Dean's list more often :D
Attending the public presentations and talks
Going to a governing council meeting :) Yeah, thinking big :D
Collaborating with students in campuses in and around Hyderabad: ISB, OUMC, NALSAR. If the talks with them donot result in collaborations atleast I would have had more friends :)
Going to class and project meetings on shorts
Invent atleast one new technology
Touring all the nearby software giants in my first year itself (Would have been much clearer on my priorities by the end of first year itself :) )
Attending a random lecture.
Experiencing athletic events (200m, 800m and others) more. [For those who saw my 400m run, it was my fastest run, gave my best :P]

These are a few things I should have done during my B.Tech life. I miss these experiences.

ps: those who are reading this post, if graduated ( or graduating soon), please do share your 'missed-experiences' too. Either academic or non academic.

Marriage between Card Game Creation and NLP Research

2010-04-23T01:28:00.000-07:00

First let me describe a bit about the card games. I classify card games into two categories:

Comparison based Card Games: In these card games each card has certain attributes. The game proceeds by comparing the attribute values of the cards - the player can decide to chose the attribute to be compared during his turn which makes the game very much enjoyable. For instance, in our childhood WWF card game, the WWF heroes are compared using various attributes such as height, weight, biceps, number of wins etc. In a more educative card game, Nutrimania, commercialized by Butterfly fields, by comparing attributes values students get to learn the nutritional facts of edible items. In order to create a comparison based game, one needs to decide on a set of objects to compared and the attributes using which one can compare them.
Relationship based Card Games: In these games there have a set of facts that the players agree on before hand. These facts specify relationship between two objects of interest (cards). After the cards are distributed players take turns and form a long chain of relationships and the one who breaks the chain loses the game. Example: Say we know the facts in biology domain. Steroid -> (is a) -> Liposoluble; Liposoluble -> (is a )-> Factor; Factor -> ( is a ) -> Molecule; Cell -> (Secretes) -> Factor; Cell->(transforms)-> Ligand; Now player1 gets the cards Liposoluble, Ligand and Factor and player2 gets cards Molecule, Steroid and Cell. They can drop cards and form the chains. If player 2 starts with Steroid he loses as the only chain that can be formed is this: Steroid -> (is a) -> Liposoluble; Now player 2 cannot extend it further. To create such games what we need is the set of facts.

Coming to the point:

What bothers me is the gap between the two communities - one which commercializes these card games and the other one, NLP research community, which excels in automatic extraction of attributes and facts from the text.

There has been a lot of research in NLP community on attribute extraction and relationship extraction problems. If the NLP community and the commercial card game creators join hands two of them together could create hundreds of educative games instead of just settling for two.

ps: The most recent publication on attribute extraction is from our very own backyard, SIEL lab, IIIT Hyderabad. An Unsupervised Approach to Product Attribute Extraction.

Santosh Raju, Prasad Pingali, Vasudeva Varma. Conference: European Conference on Information Retrieval 2009, Toulouse, France.

Winning SemEval 2010 Competition

2010-04-11T23:43:00.000-07:00

Congratulations to Siva Reddy and I Abhilash of LTRC, IIIT-H for getting the TOP TWO RANKS (in unsupervised category) in Word Sense Disambiguation (WSD) contest organized by Assoc. for Computational Linguistics. They would be going to Norway to present their award winning work at the top NLP conference in the world.

PS: Word Sense Disambiguation is an AI-hard problem which has attracted a lot of research over the past 50 years in the field of computational linguistics. Even now the state-of-art unsupervised systems perform worse than the baseline system. Their systems are the first to beat the baseline with a sizable margin. Their approach will help improve the quality of search, question answering, and machine translation.

The exact title of the contest they won is All-Word Domain-Specific Word-Sense Disambiguation as a part of SemEval 2010 contest organized by ACL-SIGLEX. They won in the unsupervised category. Top research groups from major universities all over the world participate in the contest that takes place once in three years.

This was the content of the mail circulated to the IIIT community about our achievement in SemEval 2010. I wanted to share behind the scenes activity this post.

Being habituated to working in the last minute, we delayed things till the end. Although the contest has been announced 6 months ago, we started to work on this at the very end; April 2nd was the deadline and we started on March 28th. All the work has been done in the last 5 days before the deadline. It worked for us this in this case!

Achieving this wouldn't have been possible without our enthusiastic team. Our team members are Siva Reddy, Dr. Diana McCarthy, Dr. Mark Stevenson, Prof. Rajeev Sangal and me. Mark's inputs during the formulation of our idea, Diana's constant feedback on the work during the last week before deadline and Prof. Sangal's support have helped us in achieving this feat. In the last four days we sent atleast 20 emails to Diana and to all of them she replied elaborately. It has been a wonderful experience working with Diana and Mark; we have never met them non-virtually.

The other worrying thing was my absence during the last day. I planned a holiday trip to Goa on April 2nd. That was the last day; a crucial portion of the work is still left. Having worked so far, I was not in a mood to leave the competition and go for the trip at that stage. It was a tough decision for me. I decided not to go for the trip but Siva said,

I can manage these jobs. You just go to the trip and forget about this task.

If it were not Siva, I wouldn't have gone for the trip. The discussion between us continued even while I was traveling. The last few discussions on the work happened through phone while I was in the bus to Goa. They were fruitful. Our discussions played a crucial role in the development of our system.

We were never really prepared for this competition. We were working on something else and it happened that we took part in the competition. During the last six months we didn't focus on this task at all. Siva and me have been working on another problem. We were applying algorithms from the field of Multi Agent Systems for solving the problem of Word Sense Disambiguation. We were focusing on that and recently we got our work describing preliminary results accepted at a reputed conference. It is hard to imagine what would have happened if we were just focused on the competition task?

ps: The ranks are not final.

Animal Watching and Trekking in the Forest During Night

2010-04-06T23:10:00.000-07:00

Yes, we went to into forest to watch the wildlife that were active during the night. The place is too dark and we were carrying two torch lights (more powerful than normal ones). When these lights were switched off we could not even see the person infront of us. It is not an easy task to sight animals in such a dark environment. Our guide is highly experienced and passionate about wildlife watching during nights; this has been his hobby for the past 15 years, he says. He let us experience the thrill too -- we saw deers, flying squirells, foxes and other wild birds.

What fascinated me is our guides ability to identify the exact location of the animals with these torch lights. It isn't a trivial task to identify the animals with that light. There are two reasons for this, 1) it's hard to distinguish the light reflected from the animals body and it's surroundings with this torch light 2) it's even harder when the animals are at a far off distance (more than 50 mts). All he did was scan the whole view with a beam of light, only once, and point us to a (distant) place and we find an animal over there. It looked like everything is pre-planned.

I could resist my curiosity and I walked upto him. I asked him how he was able to do it with such a precision. He revealed the trick. He said he mainly uses three features to identify the target animal in the dark.

Color of the reflected light from the eye of the animal: This is a distinguishing feature. Deers have green colored eyes, cats have blueish eyes and so on.
Distance between the eyes: Bisons have longer distance between eyes than deers.
Height of the eyes from the ground: Being experienced he knew how to use this in addition to the above features in identifying the animal.

What an insight! Being tech savvy, my immediate though was to have a mobile application which can do all this. The application works something like this: I take out my mobile and take a photo graph of complete view around me. I then simply press a button called "Identify Animals" and it does the job. It would be best if we also have a infrared viewer integrated. Such an application could give additional educative information regarding the animals ( from Wikipedia or other such source). Apart from being educative, it could be fun when used on our friends :P

ps: We went for a 4 day long trip to Goa and Dandeli. While Goa beaches are colourful, Dandeli is more involving, adventerous and thrilling; we did Rock climbling, Rafting, Canoeing and intense trekking. Overall, I enjoyed Dandeli much more than Goa.

Anti Litter Campaign: My Thoughts on Gadgets and Low versus High Tech Schools

2010-03-22T02:22:00.000-07:00

Our litter ... their MURDER .. Think before you throw.

I just returned from a tiresome Anti litter campaign held at Rajiv Gandhi Tiger Reserve at Srisailam. The idea is to make the ghat road to Srisailam litter free. Why clean it? There are atleast two reasons

Animals, especially deers and monkeys are dying as they eat the plastic covers thrown by tourists along the road .
These animals got used to eating food thrown at the road side. Because of this they are frequently appearing on the roads and getting killed in the road accidents.

WWF arranged busses from Hyderabad to Munnanur. Schools children ( 6 -10 standard) from various places in Hyderabad gathered to take up the task. We selected 6 tourists spots which are 20, 30, 40 kilometers away from srisailam. We split into groups and went to these spots. Then the task is to simply pick up all the plastic thrown by tourists alongside the road.

I got to interact with various Environmentalists. One thing I noticed is this, they keep saying X is not good, Y is results in global warming, Z is creating ecological imbalances, don't do this and that -- but they never talk of alternatives. I have heard them preaching students don't use disposable glasses made of plastic, don't use mobile phones, don't use wooden furniture, don't use paper and so on. I was vexed of hearing all that.

I asked them to talk of alternatives. The alternatives are as silly as they can get. Carry a steel glass, gunny bag (alternative to plastic bag) and furniture made of stone (our muscles permit a comfortable sitting on a stone too). I couldn't appreciate any of those solutions.

I looked for more practical, usable and affordable solutions. I thought for a while about it. Prof. Anil Gupta's talk reminded of using mobile in innovative ways. I wanted

a mobile phone that can also be used as a glass to drink water, coffee and others. Design should be in way that it doesn't increase the weight, doesn't destroy the look and it should be affordable.
a mobile phone that can be used as a carry bag. Is this a possibility? My friends said a big positive YES when I discussed the same with them.

I wanted one more gadget and idea of having this one was borne out of my frustration during the task of picking up plastic. It was too tiring in the afternoon hot sun. I wanted a gadget that can pick up only plastic which occurs in various sizes and various shapes viz: bottles, covers and carry bags would have made my life much easier. At that moment, I wished that I could design such gadget at that instance itself.

That is all I had to say about my thoughts on gadgets. I will now talk on our interactions with school kids.

We had long interactions with school children in the bus and on the field too. It was a sight to watch the hitech school children from private school and low tech school children from Govt schools.

The hitech school children were talking very fluently in English with whose fluency only a few of batch mates can match. But the irony of the situation is that they can't communicate fluently in their mother tongue. On the other hand, the low tech ones being proficient in their mother tongue, Telugu, were able to communicate fluently with villagers. During the night, we suddenly had a power cut. The high tech ones oblivious to the powercut were busy on their gadgets, some of which I haven't seen till date ( more advanced than ipods) whereas we heard shouts and joyful screams from the low tech ones 'areeii chukkalu choodandii raaaa' (have a look at the stars). Hitech kids took so much care about the food they take -- they were very much worried about their looks. They also had a natural resistance to anything new. This is not the case with the other lot; they were experimenting and trying out new things. Hitech kids were seen to be very much disciplined. However, I heard their teachers addressing these independent thinkers, often in a helpless state, "You can decide what is wrong what is right after you turn eighteen, but for now, you have to obey us". There were no such warnings to the low tech ones just because of the fact that they had fewer options to try. Infact, I would say these low tech ones had to discover their own options.

All these comparisons sounds fine and seems repeating what we all had already heard. What bothers is the fact about their school fee of hitech kids.

When I asked them the amount they pay as school fee, the answer came as a shock to me. Their fee is Rs. 1, 30, 000 per annum. It is more than my engineering college fee per annum. I wonder why they charge so much. If you don't believe me just check their website: Abhyasaa School . I asked their teachers as to why the school fee is astronomical: She explained me by saying that the infrastructure is too good. They use LCD projectors and that they have swimming pool and various other kinds of costly facilities on campus.

I always thought school education is never commercialized: I studied in a school where our fee was just Rs. 50 per month. I always had an opinion that commercialization is only at Intermediate level (IIT coaching ) and engineering level. I hate this in the sector of education. By analogy, I am looking forward to a free software kind of revolution in educational sector too. (More on this in my next post)

ps: Thanks to Kranthi for asking me to join him in this trip. Made me think on aspects to which I wouldn't have paid attention otherwise.

I don't believe in course grading

2009-06-18T03:15:00.000-07:00

This is one of my favourite posts. I have been wanting to write it since past 6 months. Finally here it is.

Note: In this post by grading i refer to the process of assigning letter grades to student scores.

I was never satisfied with the grades i got at the institute. I always felt my work deserved something better for the effort put in it. One of my observations for not getting *the* grades is the following,

In most of the courses my work always the lost the upper grade by 1 mark. Most of the times professors graded in such a way that my work was always at the letter grade boundary.

This motivated me to study the process of grading further. Prof. P.R.K. Rao once said, "Typically the youth keeps criticizing the system. They do not look for solutions".

So, I thought i should go a step further?

The solution to this is partly worked here: We asked the following question; when do we say a grading is optimal? Informally, A grading is optimal when the subtle differences in student performances are reflected in the assigned grades. To put it formally, imagine two experiments. The first one in which we select a student and inspect his score. The second in which we select a student and inspect his grade. We choose to model the problem as follows
The entropy in the distribution of grades should be equal to the entropy in the distribution of scores. i.e., the entropy of the outcomes of experiment1 should be equal to entropy of outcomes of experiment2.
What does it mean to model like this?
1. We can quantify grading quality: A grading quality factor could be defined as the difference in entropies of the above experiments. Ofcouse, the grade assignment should be monotonous. (Monotonicty: if x,y are two scores such that x> y then grade(x)> grade(y))
2. What help is the grading quality factor? Let us say we have the scores and assigned grades in a couse. We could use the above measure and say how good is the professors grading. Infact, the idea is to signal the professor that a better grading is possible. To put it in other words, we can do less injustice to students by choosing rightful grade boundaries.
3. An initial experiment and open questions: An initial experiement is conducted to see when the grading optimal. It turns out that ranking is the optimal way of grading. i.e., the number of students and grades must be equal. This can be proved easily from the definitions above. This also raises the following interesting question:
  As the number of students increase we go for ranking. For example when lakhs of students write IIT-JEE, AIEEE exams we do ranking. Then, is it justified to use the same number of grades i.e, 7(A, A-, B, B-, C, C-, D), to grade a class of 20 students as well as a class of 200/2000 students? Can we use the optimality criterion defined above to support/counter these?
4. Conclusion:
  - A grading quality factor is defined.
  - Open question1: For a given set of scores and assigned grades what is the maximal grading quality achievable?
  - Open question2: For given set of student scores what is the number of letter grade required to do optimal grading? How to characterize this?
  - If i'm going to a paper on this i would title it I don't believe in course grading.

Note: In this post by grading i refer to the process of assigning letter grades to scores. My belief is that there is a lot of subjectivity in this process. Here are few methods some professors adopt in this process

Picking k-largest gaps: If they have to assign k-grades they pick the k-largest gaps. This is not going to work in a number of cases. You can easily construct an example and verify it.
Top 33% A,A- Next 33% B,B- Next 33% C,C- and the rest D. Even this has a number of pitfalls.
Institute policies: top x% A's, y% A-'s and so on
Score > 90%=> A, >85%=>A-, >80%=>B and so on
and so on

All these have pitfalls. The same mantra is not going to work always.
I make a much stronger claim here:Any grading strategy which pre-decides the letter-grade boundaries is not going to be optimal.

Feel free to get back to me if you want to discuss more on this.

Credits: All this initially started as a dining table conversation with prudhvi. What i wrote above is a result of numerous discussions with a number of friends. I have to credit M.N.V. Kartheek especially for the discussion on initial experiment.

Am i in sports school ?

2008-10-03T03:25:00.000-07:00

Yes, i felt like i was in a sports school this semester. I have played Hockey, Football, Volleyball, Cricket, Kabaddi, Basket ball, Badminton, Table tennis in the last 2 months. It was fun participating in inter house events.

ps1: Returned from TechVista (mahabalipuram trip) today.
ps2: Badminton tournament coming up this week.

Alice: A dying man's passion

2008-09-08T00:09:00.000-07:00

I could not resist writing this post having experienced Alice, an educational software that teaches computer programming in 3-d environment. Randy paush, is the director of this project and he is been in zero gravity. Here is the link to his last lecture (if you haven't looked at it). His keynote address at SIGCSE 2008 on alice can be accessed here.

You will appreciate it better, after you give it a try. Do try it.

Linked List Problems collection

2008-07-01T09:39:00.000-07:00

Hello everyone,
I made a presentation out of the problems i encountered during my random walks across the web. I also included the problems i discussed during Data structures course with students. I was teaching assistant twice for this course :P .

Also check out the tutorials written by me to help students during c-programming course here

Recipe For Happiness

2008-06-09T23:05:00.000-07:00

While the great philosophers found happiness, in human relationships (Thomas Jefferson) , with virtue (Aristotle), in laughter (Neitzsche). Researchers have took a data driven approach to finding the dimensions of happiness by analyzing blogs.

The content analysis inspired recipe of happiness is presented below.

Ingredients
Something new
Lots of food that you enjoy
Your favorite drink
An interesting social place
Directions

Go shop for something new – something cool, make sure
that you love it. Then have lots of food, for dinner
preferably, as the times of breakfast and lunch are to be
avoided. Consider also including a new, hot taste, and one
of your favorite drinks. Then go to an interesting place, it
could be a movie, a concert, a party, or any other social
place. Having fun, and optionally getting drunk, is also
part of the recipe. Note that you should avoid any
unnecessary actions, as they can occasionally trigger
feelings of unhappiness. Ideally the recipe should be
served on a Saturday, for maximum happiness effect. If all
this happens on your birthday, even better. Bon appétit!

The original paper can be seen here

Sherlock saw the man using binoculars

2008-03-09T01:26:00.000-08:00

Recently i came across this pic. Having seen this i couldn't resist putting it up on the blog. The is an example of structural ambiguity that commonly occurs in language. I personally feel that we understand better through visual depictions rather than normal text.
Also visit this automatic text to scene convesion system. This converts natural language descriptions into images.

Shakti (Machine Translation system) And personal comments

2008-02-24T08:37:00.000-08:00

In this post i will talk a little about Machine translation system developed at IIIT-Hyderabad. And then talk about what motivates me to work in the area of Natural Language Processing( Understanding Natural Language).

Machine translation is the task of converting Natural Language text in one language to another.
The machine translation system developed at IIIT-Hyderabad Shakti can perform the following tasks.

English to Hindi
English to Telugu
English to Marathi

The accuracies are low but the system is improving.

Recenlty I watched the movie I-Robot. As for me, this movie reflected the low Natural Language Understanding capabilities of existing systems. The task to build systems with high understanding capabilities is motivating.

And if we question ourselves why the current system still don't achieve such accuracies, the following statement by late Push Singh explains it one sentence

Many AI researchers have lost touch with the original goal of building a flexible, human-level intelligence. This is partially a consequence of the need to specialize in order to make a name for yourself in the field.

But i believe that the day is not too far, when systems be able to understand language as good as humans.