Customer Feedback & Ideas for CBR Fall 09 GT

powered by UserVoice

CBR Project Ideas Forum

Log in or Sign up | Select language

This is the Fall 09 CBR Class Project Ideas page. Please write your ideas for the class project and vote for the ideas you like!

25 votes
How many votes?
Trader's ALLeY

Suggests which stocks to purchase based on an intelligent analysis/algorithms. This includes determining the probability of stock prices rising or falling, sensing current market trends/economy, digging data from a repository of previous successful buy ins, ability to customize it based on curren... more
Suggests which stocks to purchase based on an intelligent analysis/algorithms. This includes determining the probability of stock prices rising or falling, sensing current market trends/economy, digging data from a repository of previous successful buy ins, ability to customize it based on current needs. May involve some other non-trivial computations for arriving at decisions.

by CBRproj_SubhavMital | 8 comments
21 votes
How many votes?
Learning to Win Strategy Games

Given some past traces of your game playing history, this project would involve giving some CBR system suggestions while you are playing a new game. See D2 (http://home.cc.gatech.edu/ccl/3)

by sauravsahay | 2 comments
17 votes
How many votes?
Text summurization using CBR

The idea is to use pairs of <document, summary> as case base for summarizing a new document given as query. The summarization can be thought of as evaluating the similarity in structure/domain/relationship to existing cases, then using the most similar case's summary to identify the focus p... more
The idea is to use pairs of <document, summary> as case base for summarizing a new document given as query. The summarization can be thought of as evaluating the similarity in structure/domain/relationship to existing cases, then using the most similar case's summary to identify the focus points of the query document.

by deepak.zambre | 3 comments
17 votes
How many votes?
Friend Search (Find people similar to me)

Given your social profiles (such as on Facebook, delicious, or LinkedIn) and search context, this project would involve generating matches of similar profiles with case based explanation/justification.

by sauravsahay | 0 comments
14 votes
How many votes?
SceneMapper (Photo remindings)

Given a picture as a query, this application will use a case base of pictures (using flickr api or google images api) and extract 'similar' pictures based on feature similarity. (Tags: Image Recognition, classification or clustering)

by sauravsahay | 3 comments
11 votes
How many votes?
Sword-fighting with CBR

In the near future, robots are expected move beyond the structured settings of factory floors and into human populated environments, from public offices and restaurants to private homes. As advanced robots are introduced into these cluttered, uncontrolled human environments, there is an increasin... more
In the near future, robots are expected move beyond the structured settings of factory floors and into human populated environments, from public offices and restaurants to private homes. As advanced robots are introduced into these cluttered, uncontrolled human environments, there is an increasing need to design systems that interact well with humans and are able to cooperate or assist in various tasks. With this unprecedented demand for capable robots, the key challenge is the design of real-time systems that can accomplish complex physical tasks despite various sources of constraints and the presence of uncertainty.

In this work, we study the control of humanoid robots from a behavioral viewpoint, focusing on real-time behavior selection and how multiple, primitive behaviors can be combined to accomplish a complex task. We not only examine behaviors under typical constraints (e.g., environmental, postural,or task-based constraints), but also address the additional need to interact and cooperate with humans, i.e, to anticipate human actions and respond in a timely manner. We study this problem with the particularly challenging scenario of sword-based combat. It is a difficult problem since the robot must not only formulate and aggregate behaviors to accomplish physical tasks (attacks) under various constraints, but it is also operating under an adversarial domain and must anticipate and respond to moves by the opponent. Much like the game of chess, there is a strong strategic component with the additional requirements of dynamics and control. We believe this to be a powerful scenario to demonstrate real-time control since the performance of the system is easily evaluated: successful systems have swift response, aggressive attacks, and strong defense, while mistakes in reasoning have severe penalties and can be clearly identified.

I propose that we create a CBR-based system where we synthesize defensive moves by retrieving and re-using old (successful) defenses. We already have a simulation environment where we can play/record various attack/defense moves, and this project would involve building a large dataset of moves and a corresponding CBR system for defense synthesis.

by Jiuguang Wang | 0 comments
10 votes
How many votes?
Song Recommender

All the latest Web 2.0/Semantic Web stuff is heavily manifested in the music domain! Worth taking a look here: http://dbtune.org/ (and the RDF dump links - http://dbtune.org/jamendo/#dump). Also worth looking at the music ontology project - http://musicontology.com/
that drives the Pandora music ... more
All the latest Web 2.0/Semantic Web stuff is heavily manifested in the music domain! Worth taking a look here: http://dbtune.org/ (and the RDF dump links - http://dbtune.org/jamendo/#dump). Also worth looking at the music ontology project - http://musicontology.com/
that drives the Pandora music site.

This project could involve creation of a song recommendation application that used a music ontology at the core for reasoning and driving the CBR system.

by sauravsahay | 5 comments
8 votes
How many votes?
Authorship analysis

Identify the author of a piece of text/computer program based on statistical analysis of previous works. See http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.26.6652 or http://portal.acm.org/citation.cfm?id=100385 for articles on identifying authors of programs. For general stories/novels ... more
Identify the author of a piece of text/computer program based on statistical analysis of previous works. See http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.26.6652 or http://portal.acm.org/citation.cfm?id=100385 for articles on identifying authors of programs. For general stories/novels we could use N-gram analysis or Markov chains to identify authors of texts

by Akash Shah | 1 comment
7 votes
How many votes?
Fraud detection using CBR

Credit card fraud costs financial firms millions of dollars annually. Data sets for credit approval are very noisy and CBR is known to be able to handle noisy data sets efficiently. A hybrid approach using a combination of CBR/NN strategies can be used to solve this problem. Check out this paper ... more
Credit card fraud costs financial firms millions of dollars annually. Data sets for credit approval are very noisy and CBR is known to be able to handle noisy data sets efficiently. A hybrid approach using a combination of CBR/NN strategies can be used to solve this problem. Check out this paper for related work on this topic: http://home.cc.gatech.edu/ccl/uploads/45/multiple-algorithms-for-fraud.pdf . This paper outlines a hybrid approach to the fraud detection problem.The promising thing about this topic is that it is fairly unexplored using the concept of CBR, hence it is a potential prospect for a PUBLICATION at ICCBR or other CBR conferences. Please go through the above paper and pour in your suggestions about the feasibility of this idea.

by preetam.joshi | 1 comment
6 votes
How many votes?
SustainableBuildingDesigner

The main goal is to assist architects and engineers to design more sustainable buildings. Sustainable building depends on the tuning of various features such as location, spatial configuration, solar orientation, shapes, size, construction materials, main activities, energy type, etc. For that pu... more
The main goal is to assist architects and engineers to design more sustainable buildings. Sustainable building depends on the tuning of various features such as location, spatial configuration, solar orientation, shapes, size, construction materials, main activities, energy type, etc. For that purpose case retrieval will be based on extensive data sets provided by Energy Information Administration, which is a government department responsible for generating official energy statistics. Case adaptation is planned to be done within CAD environments using parametric modeling techniques.

by andres | 0 comments
6 votes
How many votes?
Automation of large application system configuration based on CBR

We plan to create a CBR system for performance driven automatic system configuration.We firstly collect the data to build the cases for a CBR system. The case include the performance data like response time, throughput and bottleneck and the configuration data. The configuration data may include... more
We plan to create a CBR system for performance driven automatic system configuration.We firstly collect the data to build the cases for a CBR system. The case include the performance data like response time, throughput and bottleneck and the configuration data. The configuration data may include attributes like, the hardware machines, web server (Apache) configuration, application server (Tomcat) configuration, cluster (CJDBC) configuration and also DB(mysql) configuration. We may also specify other attributes like the workload and number of replicas. We will then deploy the system and collect the performance data. Then we let the customer specify the case, like what is their configuration. By doing retrieval in the CBR system that we built, we can give the solution to customer, e.g., what response time or throughput the customer can expect. The customers can deploy the configuration and do the revision and reuse step of the CBR. Finally, they retained their case in the CBR system.

by kelphet | 0 comments
5 votes
How many votes?
adsense using CBR

by anonymous | 1 comment
5 votes
How many votes?
NetFlix Challenge

http://www.netflixprize.com//rules

by sauravsahay | 1 comment
4 votes
How many votes?
Planning with CBR

We have papers with CBR-based A* where you retrieve and re-use old A* plans. What hasn't been addressed in the existing literature is the use of CBR in high-dimensional motion planning cases, where standard approaches are computationally intensive and there is a big incentive to retrieve and re-... more
We have papers with CBR-based A* where you retrieve and re-use old A* plans. What hasn't been addressed in the existing literature is the use of CBR in high-dimensional motion planning cases, where standard approaches are computationally intensive and there is a big incentive to retrieve and re-use old plans. I propose that we approach this problem using a humanoid planning example (where the configuration space is huge due to the large number of joints) and developer a CBR-based rapidly exploring random tree (RRT)

by anonymous | 1 comment
4 votes
How many votes?
Context Based Search Engine

The search engine would make use of the user's search history and past preferences and user characteristics to refine and personalize search retreivals

by Sethu | 1 comment
3 votes
How many votes?
User Specific Website

Anonymous Microsoft Web Data Database: (can be found here)
ftp://ftp.ics.uci.edu/pub/machine-learning-databases/anonymous

The data is created by sampling and processing the www.microsoft.com logs.The data records the use of www.microsoft.com by 38000 anonymous, randomly-selected users. For eac... more
Anonymous Microsoft Web Data Database: (can be found here)
ftp://ftp.ics.uci.edu/pub/machine-learning-databases/anonymous

The data is created by sampling and processing the www.microsoft.com logs.The data records the use of www.microsoft.com by 38000 anonymous, randomly-selected users. For each user, the data lists all the areas of the web site that user visited in a one week timeframe.

One can auto suggest links to user based on his history and what page he is on at that moment.

by swaroopbutala | 0 comments
3 votes
How many votes?
Case based University Recommendation system

Each case is a profile of a past user whose experience is precisely stored like GRE ,TOEFL,Past History,Papers etc. And the system would recommend a university and specialization to the new user based on the past cases and it adapts to the new user from the retrieved cases in suggesting a new so... more
Each case is a profile of a past user whose experience is precisely stored like GRE ,TOEFL,Past History,Papers etc. And the system would recommend a university and specialization to the new user based on the past cases and it adapts to the new user from the retrieved cases in suggesting a new solution which might be a combination of profile evaluation criteria of the new user with that of various other users.

There can also be evaluation criteria like specific specialisation in a particular area and related work going on in the universities..

by Ramkumar | 1 comment

powered by UserVoice