Luis A. Leiva’s research while affiliated with University of Luxembourg and other places

What is this page?


This page lists works of an author who doesn't have a ResearchGate profile or hasn't added the works to their profile yet. It is automatically generated from public (personal) data to further our legitimate goal of comprehensive and accurate scientific recordkeeping. If you are this author and want this page removed, please let us know.

Publications (63)


Transfer Learning for Covert Speech Classification Using EEG Hilbert Envelope and Temporal Fine Structure
  • Preprint

February 2025

·

3 Reads

Saravanakumar Duraisamy

·

·

·

Luis A. Leiva

Brain-Computer Interfaces (BCIs) can decode imagined speech from neural activity. However, these systems typically require extensive training sessions where participants imaginedly repeat words, leading to mental fatigue and difficulties identifying the onset of words, especially when imagining sequences of words. This paper addresses these challenges by transferring a classifier trained in overt speech data to covert speech classification. We used electroencephalogram (EEG) features derived from the Hilbert envelope and temporal fine structure, and used them to train a bidirectional long-short-term memory (BiLSTM) model for classification. Our method reduces the burden of extensive training and achieves state-of-the-art classification accuracy: 86.44% for overt speech and 79.82% for covert speech using the overt speech classifier.


Examples of swiped words on a soft keyboard
ML pipeline illustration. In this work, we focus on model building
Confusion matrices for the models reported in Table 2
AutoML for shape-writing biometrics
  • Article
  • Publisher preview available

February 2025

·

2 Reads

Neural Computing and Applications

Shape-writing is a text entry method that allows users to type words on mobile devices by gliding their finger across the keyboard from one character to the next. This creates a trajectory of touch coordinates that contains rich information about the user. Previous work exploited this information to create Machine Learning (ML) models to predict demographic and behavioral targets, such as age, nationality, or handedness. However, previous work used pseudo-grid search, which is a bit tedious and rather inefficient. We show how to find better models with Automated Machine Learning (AutoML), by completely automating the architecture design process, outperforming all models reported in previous work. Our study suggests that researchers should incorporate AutoML to their training pipelines, as classification performance will likely be better than manually designing the model architecture. Taken together, our results show that it is possible to decode user’s latent information from shape-writing trajectories with higher performance than previously reported.

View access options

Text-to-Image Generation for Vocabulary Learning Using the Keyword Method

January 2025

·

16 Reads

·

·

Aaron Quigley

·

[...]

·

The 'keyword method' is an effective technique for learning vocabulary of a foreign language. It involves creating a memorable visual link between what a word means and what its pronunciation in a foreign language sounds like in the learner's native language. However, these memorable visual links remain implicit in the people's mind and are not easy to remember for a large set of words. To enhance the memorisation and recall of the vocabulary, we developed an application that combines the keyword method with text-to-image generators to externalise the memorable visual links into visuals. These visuals represent additional stimuli during the memorisation process. To explore the effectiveness of this approach we first run a pilot study to investigate how difficult it is to externalise the descriptions of mental visualisations of memorable links, by asking participants to write them down. We used these descriptions as prompts for text-to-image generator (DALL-E2) to convert them into images and asked participants to select their favourites. Next, we compared different text-to-image generators (DALL-E2, Midjourney, Stable and Latent Diffusion) to evaluate the perceived quality of the generated images by each. Despite heterogeneous results, participants mostly preferred images generated by DALL-E2, which was used also for the final study. In this study, we investigated whether providing such images enhances the retention of vocabulary being learned, compared to the keyword method only. Our results indicate that people did not encounter difficulties describing their visualisations of memorable links and that providing corresponding images significantly improves memory retention.


Fig. 1 Overview of our approach. The upper-left part (red): salience maps S(x, y; t) are computed from input video frames; then the temporal visual salience score s V (t) is computed (Sect. 2.2). The upperright part (green): from the EEG signals of multiple observers of the video, brain features f (i) t are extracted for every observer i, and from the set of these EEG features the brain-based temporal salience score s B (t) is then computed (Sect. 2.3). Middle/lower-left part (violet):
Fig. 7 EEG channel groups considered. The occipital group did not include P03 and P04
Results (Cohen's d, mean [standard error]) per frequency band
Results (Cohen's d, mean [standard error]) per feature
Brainsourcing for temporal visual attention estimation

January 2025

·

39 Reads

Biomedical Engineering Letters

The concept of temporal visual attention in dynamic contents, such as videos, has been much less studied than its spatial counterpart, i.e., visual salience. Yet, temporal visual attention is useful for many downstream tasks, such as video compression and summarisation, or monitoring users’ engagement with visual information. Previous work has considered quantifying a temporal salience score from spatio-temporal user agreements from gaze data. Instead of gaze-based or content-based approaches, we explore to what extent only brain signals can reveal temporal visual attention. We propose methods for (1) computing a temporal visual salience score from salience maps of video frames; (2) quantifying the temporal brain salience score as a cognitive consistency score from the brain signals from multiple observers; and (3) assessing the correlation between both temporal salience scores, and computing its relevance. Two public EEG datasets (DEAP and MAHNOB) are used for experimental validation. Relevant correlations between temporal visual attention and EEG-based inter-subject consistency were found, as compared with a random baseline. In particular, effect sizes, measured with Cohen’s d , ranged from very small to large in one dataset, and from medium to very large in another dataset. Brain consistency among subjects watching videos unveils temporal visual attention cues. This has relevant practical implications for analysing attention for visual design in human-computer interaction, in the medical domain, and in brain-computer interfaces at large.





Interactive Human-in-the-Loop Topic Modeling

October 2024

·

4 Reads

Broadcasting companies produce large amounts of text and audiovisual content. Extracting meaningful insights from these sources requires efficient analysis methods, which are often only palatable to data scientists. Even in large organizations, there is a critical knowledge gap: media experts manually curate work to derive insights, which is very time consuming, while engineers can use advanced data science methods but lack the domain expertise to derive key insights from the data. We propose to bridge this knowledge gap with INTEX, a human-in-the-loop interactive topic modeling application. We designed INTEX considering non-technical media experts as the main stakeholders of the application. A user evaluation shows that INTEX enables domain experts to extract and explore topics in an intuitive and efficient manner. Our work illustrates how complex applications can be made more accessible by hiding low-level details and linking these to high-level interpretations. INTEX overcomes past challenges in topic modeling, representing the future of interactive applications in this domain.




Citations (39)


... The possibility of artificial consciousness (roughly, subjective awareness in a human-designed artificial system) has been assumed within certain theoretical frameworks [1]. However, whether artificial consciousness is indeed theoretically possible, much less empirically feasible, is not self-evident, and neither proposition should be taken for granted. ...

Reference:

Preliminaries to artificial consciousness: A multidimensional heuristic approach
Awareness in Robotics: An Early Perspective from the Viewpoint of the EIC Pathfinder Challenge “Awareness Inside”

... We train reinforcement learning policies to solve the POMDP for the oculomotor control, because it has been proven to effectively address decision-making challenges in prediction of details of gaze movement [7,38,65,82]. In our detail-level implementation, we resize the input chart images to be 320 × 320 and discretize the fixation position into a 20 × 20 map. ...

EyeFormer: Predicting Personalized Scanpaths with Transformer-Guided Reinforcement Learning
  • Citing Conference Paper
  • October 2024

... Deterioration in handwriting skills, characterized by inconsistencies in size, spacing, and letter formation, indicates a progression of the disease [7]. Recent studies (e.g., [17,23]) have shown the potential of handwriting-related tasks to reveal specific cognitive deficits indicative of AD. Researchers have explored various automated methods, including drawing tasks [17], neuroimaging [31], and gait assessments [11], to capture cognitive impairments across multiple domains. ...

Ink of Insight: Data Augmentation for Dementia Screening through Handwriting Analysis

... To show users' satisfaction levels with ChatGPT for weight loss, we employed a paired sample t-test to analyze participants' satisfaction ratings of ChatGPT to influence weight loss (comparing its overall performance in generating weight loss meals and its ability to convince weight loss users). Hence, using responses we collected from the adapted satisfaction rating scale of Dubiel et al. [71] to a 7-scale (Strongly Disagree: 1; Disagree: 2; Somewhat Disagree: 3; Neutral: 4; Somewhat Agree: 5; Agree: 6; and Strongly Agree: 7), results show that there is a significant difference between users' satisfaction levels about ChatGPT's ability to influence weight loss (t(16) = 3.632, p = 0.001). Based on this finding, we further analyze users' open-ended responses, providing more context on their satisfaction ratings of ChatGPT using thematic analysis. ...

“Hey Genie, You Got Me Thinking About My Menu Choices!”: Impact of Proactive Feedback on User Perception and Reflection in Decision-making Tasks
  • Citing Article
  • July 2024

ACM Transactions on Computer-Human Interaction

... Similarly, Lopatovska et al. [23] emphasize that, depending on the task, a positive user experience (UX) may take precedence over the accuracy of outcomes. Additionally, Desai et al. [7] found that the choice of persona also affects the perceived trust, likeability, and intention to adopt. This highlights that users' engagement, satisfaction, and even trust are influenced not only by the technical performance of AI systems but also by the broader interaction process, including the agent's persona and presentation. ...

Examining Humanness as a Metaphor to Design Voice User Interfaces

... WorldCoder [140], IterClean [111], PFM+KG [94,112], [150,87] CAAFE [56] LIDA [31], InsightPlot [100] ReactionParser [77], GILT [106], BIRD [85] [37,178] R2E [65], PRISE [177], Repoformer [158] [78,169] Aug-Tree/Linear [135] [107] ...

On-device query intent prediction with lightweight LLMs to support ubiquitous conversations

... The decrease in the α frequency band and the increase in the β band were reported to be markers of visual spatial attention [16]. A link between the β band and visual attention has been shown [17], although recent findings [18] suggest that inter-frequency bands coupling may support high-level cognition functions such as affect [19]. The relationship of neural activity with visual attention was studied with simultaneous EEG-fMRI devices [20]. ...

Efficient Decoding of Affective States from Video-elicited EEG Signals: An Empirical Investigation
  • Citing Article
  • May 2024

ACM Transactions on Multimedia Computing, Communications and Applications

... Among these, "dark patterns" have emerged as a significant focus [17,32,36,52,69]. These manipulative design strategies permeate user interfaces (UI) with the aim of guiding user behavior to achieve outcomes favorable to service providers, often at the expense of users [26,47,57,66,68]. Given their widespread presence and impact on user choices and privacy, dark patterns have attracted considerable regulatory attention [14,25,48,59,76]. ...

Impact of Voice Fidelity on Decision Making: A Potential Dark Pattern?

... Technological advancements in ML and computer vision have significantly enhanced the efficiency and objectivity of remote patient monitoring systems by providing real-time data for improved care beyond conventional healthcare settings [6]. Handwriting analysis has been shown to be effective in detecting cognitive decline and changes in motor skills in AD, thus serving as an effective diagnostic tool [19]. However, despite these advances, there is no research aimed at comparing model performance using both off-line and on-line data from identical patient/healthy cohorts [5,30]. ...

The Magic Number: Impact of Sample Size for Dementia Screening Using Transfer Learning and Data Augmentation of Clock Drawing Test Images
  • Citing Conference Paper
  • December 2023