Artículos de conferencia
Permanent URI for this collection
Browse
Recent Submissions
- ItemWhose Stories are Told? Representation of English-Speaking Communities in EFL Textbooks(Red de Investigación Chilena en ELT, 2025) Pearsall, Charlotte RoseIn times of change and reinvention, communities are developing collective meanings, ideas, and practices. Such social representations, cognitions, or thinking are always related to the social and cultural tensions in our society. With this in mind, we invite the Chilean ELT community (pre and in-service teachers of English as well as academics) to share their ELT research-based representations, thinking, and voices and be part of our biannual collective discussions.
- ItemOnline combinatorial assignment in independence systems(2025) Marinkovic, Javier; Soto, Jose A.; Verdugo Silva, Victor IgnacioWe consider an online multi-weighted generalization of several classic online optimization problems called the online combinatorial assignment problem. We are given an independence system over a ground set of elements and agents that arrive online one by one. Upon arrival, each agent reveals a weight function over the elements of the ground set. If the independence system is given by the matchings of a hypergraph, we recover the combinatorial auction problem, where every node represents an item to be sold, and every edge represents a bundle of items. For combinatorial auctions, Kesselheim et al. showed upper bounds of O (log log (k)/log (k) and O (log log (n)/log (n) on the competitiveness of any online algorithm, even in the random order model, where k is the maximum bundle size and n is the number of items. We provide an exponential improvement by giving upper bounds of O (log (k)/k, and O (log (n) for the prophet IID setting. Furthermore, using linear programming, we provide new and improved guarantees for the k-bounded online combinatorial auction problem (i.e., bundles of size at most k). We show a -competitive algorithm in the prophet IID model, a -competitive algorithm in the prophet-secretary model using a single sample per agent, and a -competitive algorithm in the secretary model. Our algorithms run in polynomial time and work in more general independence systems where the offline combinatorial assignment problem admits the existence of a polynomial-time randomized algorithm that we call certificate sampler. These systems include some classes of matroids, matroid intersections, and matchoids.
- ItemExtracting and Encoding: Leveraging Large Language Models and Medical Knowledge to Enhance Radiological Text Representation(Association for Computational Linguistics (ACL), 2024) Messina Gallardo, Pablo Alfredo; Vidal, René; Parra Santander, Denis Alejandro; Soto, Álvaro; Araujo, VladimirAdvancing representation learning in specialized fields like medicine remains challenging due to the scarcity of expert annotations for text and images. To tackle this issue, we present a novel two-stage framework designed to extract high-quality factual statements from free-text radiology reports in order to improve the representations of text encoders and, consequently, their performance on various downstream tasks. In the first stage, we propose a Fact Extractor that leverages large language models (LLMs) to identify factual statements from well-curated domain-specific datasets. In the second stage, we introduce a Fact Encoder (CXRFE) based on a BERT model fine-tuned with objective functions designed to improve its representations using the extracted factual data. Our framework also includes a new embedding-based metric (CXRFEScore) for evaluating chest X-ray text generation systems, leveraging both stages of our approach. Extensive evaluations show that our fact extractor and encoder outperform current state-of-the-art methods in tasks such as sentence ranking, natural language inference, and label extraction from radiology reports. Additionally, our metric proves to be more robust and effective than existing metrics commonly used in the radiology report generation literature. The code of this project is available at https://github.com/PabloMessina/CXR-Fact-Encoder.
- ItemOn the Unexpected Effectiveness of Reinforcement Learning for Sequential Recommendation(ML Research Press, 2024) Labarca Silva, Álvaro; Parra Santander, Denis; Toro Icarte, Rodrigo AndrésIn recent years, Reinforcement Learning (RL) has shown great promise in session-based recommendation. Sequential models that use RL have reached state-of-the-art performance for the Next-item Prediction (NIP) task. This result is intriguing, as the NIP task only evaluates how well the system can correctly recommend the next item to the user, while the goal of RL is to find a policy that optimizes rewards in the long term - sometimes at the expense of suboptimal short-term performance. Then, how can RL improve the system's performance on short-term metrics? This article investigates this question by exploring proxy learning objectives, which we identify as goals RL models might be following, and thus could explain the performance boost. We found that RL - when used as an auxiliary loss - promotes the learning of embeddings that capture information about the user's previously interacted items. Subsequently, we replaced the RL objective with a straightforward auxiliary loss designed to predict the number of items the user interacted with. This substitution results in performance gains comparable to RL. These findings pave the way to improve performance and understanding of RL methods for recommender systems.
- ItemReward Machines for Deep RL in Noisy and Uncertain Environments(2024) Li, Andrew C.; Chen, Zizhao; Klassen, Toryn Q.; Vaezipoor, Pashootan; Toro Icarte Rodrigo Andres; McIlraith, Sheila A.