Journal

At our last meeting, Dr. Hassibi suggested looking into the effects of a malicious user (workers who submit wrong information during the querying phase of building a crowdsourced dataset). After Alex, our malicious user went back to the strawberry interface and inputted 30 incorrect entries, Matthew ran the classification algorithm again. The obvious outcome was that there was a decrease in accuracy with any model. Below is a video of our conversation with Dr. Hassibi:


Also in the meeting, Dr. Hassibi and I discussed a feasible outline for a convex program. Dr. Hassibi noted that an adjacency matrix A is equal to the sum of its low-rank matrix L and sparse matrix S of negative-ones where there would have been zeroes. What we planned to have finished by the end of the week is to minimize the nuclear-norm of L (which is just the trace of a matrix) and the l1-norm of S multiplied by some regularization parameter λ. The program must be subject to L = L^T (L must be symmetric), L is greater or equal to 0 (positive-semidefinite), and A = L + S. When the convex program completes, we get L-hat which we can then run spectral clustering on it. Below is the notes that were taken during the meeting:


Another factor Dr. Hassibi recommended we look into is the trend of errors over time. Before, we had calculated the number of errors with queries in intervals of 30. Instead, Matthew did it my intervals of 1. Matthew is quick to point out a sudden decrease at 66 queries where the error suddenly becomes 0 on the graph below. The reason for such is unknown and is something we plan to figure out. Dr. Hassibi, in his last remarks before he had to go to another meeting, recommended changing the number of malicious users introduced in the dataset.


Last weekend, my family and I snuck out of our house to visit our house in Laguna for the natural phenomenon known as the bioluminescent waves. It was an hour-long drive to get down there because of how open the highways were. Of course, during our time there we socially distanced ourselves from everyone even though the beach was practically empty. We stayed until 9:00 and then back home with a binder of photos. Below is a video on what "blue waves" are about:


Comments

Post a Comment

Popular posts from this blog

Journal

Image
The other group, which included me, worked on creating an adjacency matrix through triangle queries. We got 150 queries in the Airtable database. The dataset was processed on Matlab making adjacency matrices for 30, 60, 90, 120, and 150 queries. Below is a visualization of the adjacency matrix of the dataset after 150 queries. Yellow represents an edge, teal represents no edge, and blue represents an edge not observed. Matthew did spectral clustering on the five adjacency matrices. Below is a plot of the graph of 30 queries. From the graph, we derive two clusters and 4 outliers. Matthew notes that spectral did so well that he started to become suspicious of the integrity of the dataset. After reviewing all the results, he notes that there is some positive correlation with performance and queries. Our goal for the next meeting is to get another dataset and compare its results with the strawberry one. For me, my idea for the next few weeks is to create and improve a conv...
Read more

First Month

Image
The first four weeks of this Caltech Stem class have been interesting, to say the least. Much of the material we went into have been all over the place with various objectives. We started off with teamwork exercises, watching college students present their research, replicating one for ourselves, writing a concept map, and going to Caltech to meet Dr. Hassibi. We had two, teamwork workshops for the first two weeks. One of which was the dominoes problem where each student was given two dominoes and we were instructed to get three dominoes per person by exchanging dominoes for chips and chips for dominoes. Instructions were very vague other than that everyone must have a domino to "survive" a round. We struggled in the first few attempts but through several days of discussion and trial & error, we were able to find a way to accomplish our goal. Some key underlying goals of this exercise were the importance of leadership, communication, and open-mindedness...
Read more

Filler Journal - 10/18/19 - 11/15/19

Image
     On Thursday, November 14th, I ran into a multitude of issues. All of which was because of pent up anxiety and downright disappointment. It all began in my debate class. I had a tournament coming up soon and my case was barely finished. I know I could finish it late in the day but I felt the need to do it during class. I was compelled to do so since I wanted to do well but I ended up not because I had a math test one period away. For the rest of the class, I did a last-minute cram session but I guess that was bad luck. Right before the test, my teacher, Mrs. Linton, said that if we had 6 minutes left, we "should get on the calculator section ASAP." Judging from her statement, I wrongfully judged that I would need around 10 minutes since I thought she implied that the calc section would take 6. Instead, I really needed much more time. I ended up doing extremely poor and got disappointed. During my "alone" time of thinking about what could I have done to do better...
Read more