Weekly Report

Posted 2012.12.24 03:56


2013-01-21 ~ 2013-01-25

Last Week

1. Gaussian Process Regression under Localization Uncertainties

1. Implemented Jacobi over-relaxation(JOR) method and Discrete-time average consensus(DAC) method

2. Writing Draft paper

3. paper survey

2. Intention Learning

1. Compared result with kmeans and svm

[Final Prediction Reuslt using Deep Learning]

Elapsed time is 4.066772 seconds.

1. Prediction : 97.02 %

2. Precision  : 87.94 %

3. Recall      : 92.39 %


[Final Prediction Reuslt using SVM]

Elapsed time is 106.961909 seconds.

1. Prediction : 96.25 %

2. Precision  : 98.69 %

3. Recall      : 75.49 %


[Final Prediction Reuslt using kmeans]

Elapsed time is 0.789430 seconds.

1. Prediction : 69.91 %

2. Precision  : 15.42 %

3. Recall      : 23.36 %



This Week

1. Gaussian Process Regression under Localization Uncertainties

1. Write draft paper

2. Intention Learning

1. Write draft paper

2013-01-14 ~ 2013-01-18

Last Week

1. Gaussian Process Regression under Localization Uncertainties

1. Tested performance of Distributed GP algorithm with randomly positioned agents 







2. Intention Learning

1. Tried to add convex constraints to the existing learning rule of deep belief network. However, the result was not meaningful.


3. Connectivity Control Algorithm 

1. Modifying term paper. 



This Week

1. Gaussian Process Regression under Localization Uncertainties

1. Start writing draft paper. 


2. Intention Learning

1. Read existing papers related to Intention Learning, Smart house, and Sensor, activity prediction. 


3. Connectivity Control Algorithm 

1. Modify term paper




2013-01-07 ~ 2013-01-11

Last Week

1. Gaussian Process Regression under Localization Uncertainties

1. MAP-GP vs QDS2



2. Reading distributed algorithms. 

3. Implementing JOR algorithm. 

2. Intention Learning 

1. Testing the performance of intention learning(predicting) algorithm using DBN

2. Used actual binary sensor data-set. 

3. Result

Average Prediction rate:      93.7 % <= 이 값이 잘나오는 이유는 센서가 activate안되는 경우도 포함되기 때문이다. 

------------------------------------

Average Precision rate:       81.1 % <= 모든 데이터에 대해서

Average Recall rate:          96.1 %

------------------------------------

Average Real Precision rate:  68.4 % <= activate이 일정값 이상되어서 DBN으로 예측했을 때

Average Real Recall rate:     95.4 %

------------------------------------

Precision at newly activated: 76.1 % <= 의미있는 데이터! activation이 일정값 이상이고 센서가 새롭게 activate되었을 때 맞출 확률

Recall at newly activated:    52.5 %

3. Seminar in Aerospace Engineering

1. Wrote a paper about connectivity control algorithms. 


This Week

1. Gaussian Process Regression under Localization Uncertainties

1. Finish implementing JOR algorithm. 

2. Test the performance of distributed algorithm. 

2. Intention Learning

1. Thinking about combining convex optimization and DBN. (similar to Sparse DBN)

2. Compared the performance between DBN and other algorithms. 

3. Multi-Robot Coordination 




'Thoughts > Dear Diary' 카테고리의 다른 글

마태복음 1장 1절 - 박영선 목사님  (0) 2014.12.24
나이가 드나보다.  (0) 2013.05.26
Weekly Report  (0) 2012.12.24
2012-2 시간표  (0) 2012.09.04
과연 유일신신앙은 인류 문명의 적인가?  (0) 2012.04.17
외로워서인가보다.  (0) 2012.04.03
« PREV : 1 : ··· : 29 : 30 : 31 : 32 : 33 : 34 : 35 : 36 : 37 : ··· : 55 : NEXT »