Date: 03/31/11 | Title: Project teams and topics |
Project teams and topics
These topics can change.
Jialong Zhang and Wei Li
URL maliciousness detection: Investigating relevant features
Raymond Lin and Francisco Vides
Coevolution and predator and prey using NEAT
Gopal Krishna Kedia and Saurabh Mahta
Document classification: Exploiting session based implicit links
Sumanth Reddy Kommidi and Santosh Kumar Eppalapally
Interest classification and recommendation system in Twitter
Ben Fine and Joshua Pschel
Distinguishing Between Finger-Sketched Shapes versus
Text Using Higher-Order Entropy
Hutson Betts and Spencer Huang
Determining most informative features and minimum amount
of information neeeded from movie reviews
Nakul Anil Ingley and Haokai Lu
Active vision with neuroevolution
Sampath Jayarathna and Radhika Kashyap
Community Recommender Service (CRS)
Angela Faragasso and Flavia Schiavo
Character recognition
Santosh Kumar Vanaparthy
Maximizing information transmission over finite state
erasure channel with memory
Mohammadali Tarrahi and Amin Rasekh
Efficient probability density function estimator for
Bayesian optimal classification: Application to Abalone
benchmark data set
Subhadeep Chakraborty and Varun Raj
Tag recommendation for text documents
Shuhan Xu and Mingqu Yue
Genetic algorithms for spelling correction
Longfei Zhang and Shuai Ye
Detecting beauty in color combination
Jonathan Hall and Hone-hoe Kim
Automatic title generation from text documents of different type
|
Date: 03/24/11 | Title: Xbraitenberg package debugging |
To fix compilation problem in the xbraitenberg package,
change the beginning part in vector.hh as follows:
#ifndef VECTOR_HH
#define VECTOR_HH
#include
#include
#ifdef HAVE_NEW_H
#include
#else
//static inline void *operator new(size_t, void *v) { return v; }
#endif
#include
...
Once you get it compiled, try
./xbraitenberg -l l2a l2b v2b v2b v2b v2b v2b
|
Date: 03/23/11 | Title: Project Instructions |
Code-base
- Backpropagation: backprop-1.6.tar.gz (C++ code -- unix)
- Neuroevolution: ga.m (Octave code)
- SIDA: sida-nat.tar.gz (Octave code)
Final miniproject details
1. Pick a code-base
- YC's backpropagation code (in c++)
- YC's neuroevolution code (in octave)
- YC's SIDA code for sensorimotor semantics (in octave)
- Your own code
- Reinforcement learning code (from homework)
- Other
- A third-party open source: NEAT, etc.
2. Formulate your research problem
- Pick task
- Pick a task related to your own research area.
- Locate data set (if needed): see, e.g., UCI ML repository
- Design experiments
- This is perhaps the most important!
- Just comparing existing learning algorithms on
a data set is not meaningful.
- Design experiments to discover new insights about
data.
Example:
- Discover new structure in data (using dim reduction, etc.)
- Discover dependencies among features
- Novel forms of data (connectivity, etc.)
- etc.
- Critically evaluate commonly accepted or implicit
assumptions.
Example: document classification
1. single document, natural language processing needed.
2. single document, counting word cooccurrence enough.
3. need to consider not just single document but
documents linked to it.
4. etc.
- Derive new learning algorithms
Example:
- semisupervised learning
- active learning
- ...
3. Proposal
- Team members: max 2 per team. Individual projects are allowed only
under special circumstances.
- What is the research problem?
- Why is it important/interesting?
- What are other people's approaches?
- What are other people's assumptions? (optional)
- What are the limitations of those approaches?
- What are the problems with those assumptions? (optional)
- What is your approach?
- How will you relax the assumptions? (optional)
- What experiments will you do?
- What data set will you use (if applicable)
- What code base will you use
- What kind of experiments will you do with the data/code?
- What are the expected results?
- What do you expect to achieve by relaxing the assumptions? (optional)
- Submit by 3/29 3/31 (Thu), in class. Hardcopy.
4. Final report
- 4-5 page, single space report. One column or double column.
- Conference style (e.g., IEEE conference style).
|
Date: 03/03/11 | Title: Generating random events |
How to generate probabilistic events:
i. Given choices X and probability p(X).
ii. Choose from choice
c = ceil(rand*cardinality(X))
iii. Generate random number between 0 and 1.
r = rand;
iv. if (r<p(X=c))
accept choice c
else
reject and go to step ii.
|