Objective:

The goal of this project is to construct a model (built using nnsim) which accurately explains and predicts aspects of typing skill acquisition.

Summary:

This project consists of: (1) building and using a platform to collect test data of keyboard skill acquisition, (2) constructing a model of the keyboard learning process, (3) tuning model parameters to fit data and assessing their accuracy in modeling differing tasks, and (4) using observations about model behavior to assess current keyboard learning methodologies.

Model demonstration:

This project has two key deliverables: the keyboarding task test platform and the keyboarding model.

The keyboarding task platform will be a web-based tool which displays a sequence of symbols, collects user key press responses, and (optionally) presents the user with keying accuracy feedback. This will be constructed using an HTML5/Javascript front-end task communicating results to a remote database via AJAX. The advantage of this approach is that tasks can be completed remotely on any computing platform by many users simultaneously.

The initial keyboarding model will be constructed using the NNSim framework. NNSim will be used to play with different neural network topologies, update rules and functionally important model parameters. The behavior of this model will be tuned with intuition about key learning behavior. Time permitting, a version of the model may be constructed which can interact with the keyboarding task platform, modeling the test subject’s current knowledge state in real time. Depending on the flexibility and update rule required by the model, pre-built neural network packages such as brain.js may be used.

Relevant literature:

Milestones:

11/01:
Client-side portion of keyboarding task able to prompt and record key-presses (timing and key). At least one attempt at model constructed. Update: Simple model laid out using NNSIM, investigation into client-side task framework begun.
11/08:
Server-side portion of keyboarding task built, able to collect task data. NNSim model complete and some experimentation has been performed with it. Update: Still working on client-side portion of task framework, now displaying stimuli, collecting keystrokes and timings.
11/15:
Keyboarding task platform maturing, begin rounds of data collection for varying tasks. Model being tweaked as task data comes in. Update: Began working on server-side of task, tweaking task. Experimenting with NNSIM training functionality and using grouped IAC units. Possibly meeting next week for help training in NNSIM.
11/22:
New tasks conceived of and completed. Model being tweaked to accommodate different task results. Update: Found useful paper
11/29:
Model performance assessed, begin experimenting with model to make predictions about keyboard learning. Begin building version of model suitable for integration with keyboarding task platform.Completed.
12/06:
Perform experiments with task platform, further assess model. Prepare presentation.Completed.