Browsing by Subject "Natural Language Processing"

Now showing 1 - 2 of 2
  • Thumbnail Image
    Item
    Attacking ELECTRA-Small: Universal Adversarial Triggers for Reading Comprehension
    (2023-12-14) Gupta, Alok
    New question answering computer programs are pretty good at their job. They perform well when tested on certain datasets, like SQuAD (Rajpurkar et al., 2016) However, these datasets might not fully represent the challenges that come up in real-world situations where understanding what you're reading is important.Even the most advanced models can be affected by disruptions and adversarial text that is meant to decrease a model’s performance. In our paper, we gener- ate adversarial text examples to lower accuracy for the task of reading comprehension testing against the ELECTRA-Small model (Clark et al., 2020) trained on SQuAD. We suggest using a method called "beam search" to create what we call "universal triggers." We then test these triggers on different parts of the SQuAD dataset, with different vocabulary lists. The adversarial trig- gers produced by our beam search are able to effectively reduce a model’s accuracy on the SQuAD dataset by 33% for ’Why’ questions.
  • Thumbnail Image
    Item
    Trust filter for disease surveillance : Identity
    (2016-12) Lin, Guangyu; Barber, Suzanne
    A flexible and extensible mobile application was delivered for evaluation and optimal inclusion of NextGen (Next Generation) data sources into biosurveillance for early detection, situational awareness and prediction. We present trust analysis of NextGen data sources to increase data confidence. One of the trust filters is the Identity filter, which helps us determine the degree of separation between the sender and the subject of a sentence. In this thesis, the author presents the definition of Identity. To help us distinguish different degrees of separation, the author uses relation distance along with a family tree to weight different relationships. Then the author compares a discriminative algorithm and a generative algorithm to calculate a user's Identity score. In the end, the author concludes that it is a good choice to apply a binary classification algorithm combined with a Natural Language Processing algorithm because of the trade-off between accuracy and runtime complexity.