MOCAS

Question 1

What is the Problem/Motivation?

Answer 1

They proposed a multimodal dataset for objective cognitive workload assessment on simultaneous tasks with closed circuit television tasks for better applicability to real word scenarios

Question 2

Study Overview

Answer 2

The aim of this study was to develop a multimodal dataset for objective cognitive workload using tasks that are applicable to real world scenarios.

Objective cognitive workload is the measurable mental effort or cognitive work expended while performing a task.
In this study, the objective cognitive workload is measured through physiological and behavioral signals.

During the experiments, the participants were asked to monitor video feeds from multiple robots and identify specific objects.

Their physiological and behavioral data was simultaneously collected through (emotiv and empatica (phys)) and (Intel Realsense D435i (facial recognition)) sensors.

Following the experiments,
In the analsis phase, they calculated the pearsons correlation coefficient find the relationship between the particpipants personality traits and the average of their physiological and behavioral data.

• In addition, LSTM and LF-LSTM models are used for unimodal and mulitmodal classification of cognitive load.
• LSTM models excel in capturing temporal patterns by learning from the sequential data.
• LF-LSTM provides an effctive way to intgerate different modalitis by procssing the featres from each modality independently before combinig them. Hence the LF (Late Fusion).

Question 3

Correlation between personality traits and physiological and behavioral signals

Answer 3

Studies have shown that different personality traits can generate physiological and behavioral signals with different features.
- The calculates pearsons correlation coefficient to find the relationship between the participant’s personal traits and the mean of their physiological and behavioral data.

• Resulted in a moderate positive relationship

Question 4

LSTM

Answer 4

A deep learning model that excels in capturing temporal patterns

Question 5

LF-LSTM for multimodal fusion classification

Answer 5

Late-Fusion long short term memory networks

the benefit of using LF-LSTM is that:
features extracted from each modality are processed independently before they are combined, capturing the unique characteristics of each modality before combining them.

Question 6

Model Evalutions

Answer 6

• Did well with trial independent evaluation
• struggled with subject independent evaluation, suggesting that the model struggled to account for individual differences it was not trained on, probably
  due to the fact their subject size was so small, having only 21 participants.