Skip to content

1. Meeting Notes 09|25

Jump to bottom
Daniel Obeng edited this page Nov 8, 2017 · 1 revision

Attendees: Daniel, Hannah, Tarekul & Socratis

Meeting Purpose: Idea Generation & Evaluation

Summary: During this meeting, our team came together to discuss possible ideas

IDEA 1: Cybersecurity

A system that is able to flag users on a network for malicious activity. The software will analyze simulated data and learn from that data. Then, in a real system, it will be able to flag a user if he might be involved in some malicious activity with some accuracy (between 70 - 80%)

Challenges:

  • A virtualized network environment is needed in order to test the system which will require some sort of hardware or cloud service.
  • Data might need a lot cleaning up before it can be utilized
  • Might require a lot of work to create a working product

IDEA 2: Intrusion System

Intrusion system to break into a network - doesn't seem feasible, discarded

IDEA 3: Best Route

Get data from Taxi and Limousine commission. Visualize it on a map and find the best possible routes for uber/lyft pickups. The goal is to display where the most business is so drivers and frequent these routes and possibly get more business.

Challenges:

  • It doesn't seem very difficult. The solution will be the creation of an algorithm that any programmer can write.

IDEA 4: Voice Recognition System

Machine learning voice recognition system; where a person speaks and it analyzes and understands its voice. Then be able to identify the voice of person among other voices later on.

NOTE: Professor commented that this could be a possible project to work on

Suggestions by Professor:

What about a voice recognition that identifies male and female voices?

PROFESSOR'S IDEA - Combine voice recognition with Security

Detect from voice recognition - that a person is a threat. Based on what this person is saying.

MORE DETAILS:

If we already know the voice of the person as a malicious individual, then we flag it as a threat. If not, then we analyze the words he's saying and then determine if the user is a threat. (Some people may say words and they are simply kidding based on the context)

We collect some data and then determine whether a person is malicious or not.

He said it could be commercially interesting. We can not only.

Application, for salesmen, they would like to know if a user wants to buy something

There are some software already available to recognize voice. We can use it but we have to add something interesting on top of it.

NOTE: If you read a scientific paper for this project, put it down in your record and show the class that

OTHER IDEA

Voice - combine voice with video. Face - we can easily know where the face is in a video using some readily available package Video, if the person is saying something bad. Able to identify faces of people, we can feed it data to recognize people.

Based on face, based on some face - we can determine what the person is saying, what emotion he's saying it with to determine if he's a threat.

We synchronize the voice with the face and then we analyze the face and the voice and see if the person is threatening or if it is a funny or friendly message.

Based on voice and the face we can analyze what the person is saying.

In our project, assume that we have both inputs are constant.

PAST PROJECTS

Last year, one team worked on voice. One worked on gesture but it wasn't a sucessful project. The best team is when the classify twitter sarcastic comments, they can achieve about 90% accuracy.

Shoot for 80% accuracy. Don't be too hard, unless you want to cheat. Maybe manipulate data a little bit.

TODO

  1. Github or gitlab report to be linked to the course web
  2. Brainstorming presentation next monday (not graded)
  3. Start to push meeting records in your repo
  4. Check the commits to looks

Clone this wiki locally