Article
Computer Science, Artificial Intelligence
Luca Oneto, Sandro Ridella, Davide Anguita
Summary: Recent research has shown that machine learning models, especially deep learning models, can be easily manipulated by attackers through imperceptible modifications of input data. This discovery has led to the development of adversarial machine learning, where new attack and defense methods are continuously developed, resembling the field of cybersecurity. In this paper, we demonstrate that inducing models from data less prone to manipulation can actually benefit the assessment of their generalization abilities. We present both theoretical perspectives using state-of-the-art statistical learning theory and practical examples.
Article
Computer Science, Artificial Intelligence
Gonzalo A. Aranda-Corral, Joaquin Borrego-Diaz, Juan Galan-Paez
Summary: This paper discusses the changes in concept mining capability when reconsidering the hypothesis class, from the perspective of the Three-Way Decision (3WD) paradigm, and introduces new versions of the Vapnik-Chervonenkis dimension. It aims to analyze the influence of 3WD techniques in the Concept Learning Process.
INTERNATIONAL JOURNAL OF MACHINE LEARNING AND CYBERNETICS
(2022)
Article
Computer Science, Information Systems
Linghui Kong, Wenzhong Yang, Fuyuan Wei
Summary: Implicit sentiment analysis differs from traditional text sentiment analysis by not relying on emotional words as emotional clues and having vaguer expressions. Identifying implicit emotions is more difficult as it requires a deeper understanding of the context, even when emotional words are absent. Researchers focus on context feature modeling and developing sophisticated feature extraction mechanisms instead of starting from the emotional perspective. Enhancing the difference in emotional features of text samples is an intuitive method to address this challenge. The proposed supervised contrastive learning (SCL) method enables the model to conduct contrastive learning based on emotion labels even while training on weak emotion features, enhancing implicit emotion discrimination. The use of a straightforward context feature fusion method (bi-affine) over a more complicated context feature modeling approach is applied to improve implicit emotion classification ability.
Article
Statistics & Probability
Man Fung Leung, Yiqi Lin, Nicolas Wicker
Summary: This paper corrects the learning risk lower bound in the non-realizable case provided by Anthony and Bartlett in 1999. The main contribution is the technical corrections made to their proof, including correcting the lemma used and adapting the lower bound proof itself.
COMMUNICATIONS IN STATISTICS-THEORY AND METHODS
(2023)
Article
Computer Science, Artificial Intelligence
Jian Li, Yong Liu, Weiping Wang
Summary: In this study, we derive novel semi-supervised excess risk bounds for vector-valued learning by using local Rademacher complexity and unlabeled data from both kernel and linear perspectives. The derived bounds are sharper than existing ones and the convergence rates are improved. Based on our theoretical analysis, we propose a general semi-supervised algorithm that incorporates both local Rademacher complexity and Laplacian regularization for efficiently learning vector-valued functions. Extensive experimental results show that the proposed algorithm outperforms compared methods, aligning with our theoretical findings.
PATTERN RECOGNITION
(2023)
Article
Computer Science, Artificial Intelligence
Falco J. Bargagli Stoffi, Gustavo Cevolani, Giorgio Gnecco
Summary: This article discusses the relationship between simplicity and truth, as well as how to balance simplicity and accuracy in supervised machine learning. The study found that in certain situations, excessive pursuit of simplicity actually slows down the learning process. These results have important implications for both machine learning and the philosophy of science.
MINDS AND MACHINES
(2022)
Article
Computer Science, Artificial Intelligence
Qizhou Sun, Yain-Whar Si
Summary: In this paper, a novel framework named SACRL-AF is proposed to address the issue of incomplete fulfillment of buy or sell orders in algorithmic trading. The framework includes an action feedback mechanism for notifying the actor about dealt positions and an approach of using dealt positions as labels for supervised learning. Based on the SACRL-AF framework, two reinforcement learning algorithms, SDDPG-AF and STD3-AF, are proposed and achieve state-of-the-art profitability in experimental results.
APPLIED INTELLIGENCE
(2023)
Article
Computer Science, Artificial Intelligence
Xunlin Zhan, Yuan Li, Xiao Dong, Xiaodan Liang, Zhiting Hu, Lawrence Carin
Summary: This paper introduces the elBERto framework, which improves model's ability to leverage rich commonsense in context through five self-supervised tasks. Experimental results show that elBERto outperforms other methods in challenging questions and achieves substantial improvements.
KNOWLEDGE-BASED SYSTEMS
(2022)
Article
Computer Science, Artificial Intelligence
Yanyan Wang, Qun Chen, Murtadha H. M. Ahmed, Zhaoqiang Chen, Jing Su, Wei Pan, Zhanhuai Li
Summary: This paper proposes a supervised GML approach for ATSA, which effectively exploits labeled training data to improve knowledge conveyance. The approach utilizes binary polarity relations between instances to enable supervised knowledge conveyance. By modeling detected relations as binary features in a factor graph, the proposed approach fulfills knowledge conveyance. Experimental results demonstrate that GML outperforms pure DNN solutions when collaborating with DNN for feature extraction.
TRANSACTIONS OF THE ASSOCIATION FOR COMPUTATIONAL LINGUISTICS
(2023)
Article
Computer Science, Artificial Intelligence
Sijie Mai, Ying Zeng, Shuangjia Zheng, Haifeng Hu
Summary: The wide application of smart devices has enabled the use of multimodal data, but training networks with cross-modal information is still challenging due to modality gap. Additionally, the learning of inter-sample and inter-class relationships is often neglected. To address these issues, we propose HyCon, a framework for hybrid contrastive learning, which can explore cross-modal interactions, learn inter-sample and inter-class relationships, and reduce the modality gap. Our method outperforms baselines on multimodal sentiment analysis and emotion recognition.
IEEE TRANSACTIONS ON AFFECTIVE COMPUTING
(2023)
Article
Computer Science, Artificial Intelligence
Yuki Odaka, Ken Kaneiwa
Summary: To represent emotions in computer applications, Russell's circumplex model is used, which classifies emotions based on valence and arousal. This paper presents a method for predicting arousal levels using a block segmentation vector based on semi-supervised learning. The method involves generating word vectors based on inverted indexes and dividing the corpus into blocks. Experimental results show that this method outperforms previous methods in SentiWordNet and SocialSent.
APPLIED SOFT COMPUTING
(2023)
Article
Computer Science, Artificial Intelligence
Christian Antic
Summary: Analogy-making is essential for human and artificial intelligence, and this paper introduces an abstract algebraic framework for handling analogical proportions. By showing that analogical proportions can be embedded into first-order logic and are compatible with structure-preserving mappings, the plausibility and applicability of the proposed framework are demonstrated. This paper marks an important first step towards a theory of analogical reasoning and learning systems, with potential applications in common sense reasoning and computational learning and creativity.
ANNALS OF MATHEMATICS AND ARTIFICIAL INTELLIGENCE
(2022)
Article
Computer Science, Software Engineering
Steve Kommrusch, Martin Monperrus, Louis-Noel Pouchet
Summary: This article addresses the problem of automatically synthesizing proofs of semantic equivalence between two programs. A neural network architecture based on a transformer model is proposed to generate proofs of equivalence using semantics-preserving rewrite rules. The system achieves a 97% proof success rate on a dataset of 10,000 pairs of equivalent programs.
IEEE TRANSACTIONS ON SOFTWARE ENGINEERING
(2023)
Article
Mathematics
A. Iosevich, B. McDonald, M. Sun
Summary: This article investigates the Vapnik-Chervonenkis dimension of the classifier set 9-3t(E), where E ⊆ Fq3, and proves that the dimension is 3 when |E| ≥ Cq4. This implies that for sufficiently large subsets, the dimension is the same as the Vapnik-Chervonenkis dimension of Fq3.
DISCRETE MATHEMATICS
(2023)
Article
Political Science
Derek Beach, David Schaefer, Sandrino Smeets
Summary: This article examines the role of analogical reasoning in epistemic learning and develops a two-stage model that transfers core causal lessons while considering contextual differences. Evidence from a Banking Union case study shows that analogical reasoning played a significant role in shaping policy debates. The conclusion calls for policy analysts to pay more attention to how comparisons with the past influence policy discussions.
POLICY STUDIES JOURNAL
(2021)
Review
Computer Science, Theory & Methods
Temitayo Olugbade, Marta Bienkiewicz, Giulia Barbareschi, Vincenzo D'amato, Luca Oneto, Antonio Camurri, Catherine Holloway, Marten Bjorkman, Peter Keller, Martin Clayton, Amanda C. De C. Williams, Nicolas Gold, Cristina Becchio, Benoit Bardy, Nadia Bianchi-Berthouze
Summary: This paper presents a catalog of 704 open datasets that can be valuable to researchers searching for secondary data. The datasets are analyzed and reviewed under themes such as human diversity, ecological validity, and data recorded. The resulting 12-dimensional framework can guide researchers in planning the creation of open movement datasets.
ACM COMPUTING SURVEYS
(2023)
Article
Computer Science, Artificial Intelligence
Seham Basabain, Erik Cambria, Khalid Alomar, Amir Hussain
Summary: An increasing number of studies are using pre-trained language models to tackle few/zero-shot text classification problems. However, most of these studies fail to consider the semantic information embedded in the natural language class labels. This work demonstrates how label information can be leveraged to enhance feature representation in input texts, particularly in scenarios with scarce data resources and short texts lacking semantic information like tweets. The study also shows the effectiveness of zero-shot implementation in predicting new classes across different domains, achieving high accuracy in Arabic sarcasm detection.
Article
Computer Science, Artificial Intelligence
Kai He, Yucheng Huang, Rui Mao, Tieliang Gong, Chen Li, Erik Cambria
Summary: This paper proposes a virtual prompt pre-training method that incorporates the virtual prompt into PLM parameters to achieve entity-relation-aware pre-training. The proposed method provides robust initialization for prompt encoding and avoids the labor-intensive and subjective issues in label word mapping and prompt template engineering.
EXPERT SYSTEMS WITH APPLICATIONS
(2023)
Article
Computer Science, Artificial Intelligence
Mostafa Amin, Erik W. Cambria, Bjorn Schuller
Summary: ChatGPT demonstrates the potential of general artificial intelligence capabilities and performs well across various natural language processing tasks. This study evaluates ChatGPT's text classification abilities for affective computing problems including personality prediction, sentiment analysis, and suicide tendency detection. Results show that task-specific RoBERTa models generally outperform other baselines, while ChatGPT performs decently and is comparable to Word2Vec and BoW baselines. ChatGPT exhibits robustness against noisy data, outperforming Word2Vec in such scenarios. The study concludes that ChatGPT is a good generalist model but not as specialized as task-specific models for optimal performance.
IEEE INTELLIGENT SYSTEMS
(2023)
Article
Computer Science, Artificial Intelligence
Mauajama Firdaus, Asif Ekbal, Erik Cambria
Summary: This research proposes a multilingual multitask approach that improves intent accuracy and slot filling for three different languages. The experimental results show an improvement in both tasks for all languages.
INFORMATION FUSION
(2023)
Article
Engineering, Civil
Andrea Coraddu, Luca Oneto, Shen Li, Miltiadis Kalikatzarakis, Olena Karpenko
Summary: In this paper, a three-step approach is developed for the optimal design of stiffened panels accounting for the ultimate limit state due to welding residual stress. State-of-the-art analytical approaches coupled with data-driven nonlinear finite element methods surrogates are used to find new optimal designs. The results obtained from optimizing a series of parameters of a commonly used stiffened panel geometry will support the authors' novel approach.
ENGINEERING STRUCTURES
(2023)
Article
Computer Science, Artificial Intelligence
Luca Oneto, Sandro Ridella, Davide Anguita
Summary: This century has witnessed a significant increase in investments in Artificial Intelligence (AI) and particularly in (Deep) Machine Learning (ML) in both public and private sectors. This has led to breakthroughs in solving complex real-world problems, but the increasing complexity of these breakthroughs has made it difficult to understand their fundamental mechanisms. Researchers are now questioning the need for a new theoretical framework to catch up with this complexity. One particular mechanism that is not well understood is over-parametrization, which refers to the ability of certain models to improve their generalization performance even when the number of parameters exceeds the interpolating threshold.
Review
Engineering, Marine
Giulia Cademartori, Luca Oneto, Federica Valdenazzi, Andrea Coraddu, Andrea Gambino, Davide Anguita
Summary: The prediction of ship motions and quiescent periods is crucial for the maritime industry to improve safety and efficiency of various marine operations. This review categorizes and examines the existing ship motion and quiescent period prediction models into physical, data-driven, and hybrid models. The advantages, disadvantages, open problems, and future perspectives of these models are discussed.
Article
Computer Science, Artificial Intelligence
Dazhi Jiang, Runguo Wei, Jintao Wen, Geng Tu, Erik Cambria
Summary: Emotion recognition in conversations has wide applications in various fields. We propose an AutoML strategy based on emotion congruent effect to select suitable knowledge and models, and effectively capture context information and enhance external knowledge in conversations.
IEEE TRANSACTIONS ON AFFECTIVE COMPUTING
(2023)
Editorial Material
Computer Science, Artificial Intelligence
Frank Xing, Bjoern Schuller, Iti Chaturvedi, Erik Cambria, Amir Hussain
Summary: Neural network-based methods, such as word2vec and GPT-based models, have achieved significant progress in AI research, especially in handling large datasets. However, these methods lack in-depth understanding of the internal features and representations of the data, leading to various problems and concerns.
IEEE TRANSACTIONS ON AFFECTIVE COMPUTING
(2023)
Article
Computer Science, Artificial Intelligence
Kai He, Rui Mao, Tieliang Gong, Chen Li, Erik Cambria
Summary: The study proposes a meta-based self-training method for aspect-based sentiment analysis (ABSA). By generating pseudo-labels and controlling convergence rates, the method improves model performance and accuracy in fine-grained sentiment analysis.
IEEE TRANSACTIONS ON AFFECTIVE COMPUTING
(2023)
Article
Computer Science, Artificial Intelligence
Tan Yue, Rui Mao, Heng Wang, Zonghai Hu, Erik Cambria
Summary: Sarcasm detection is a challenging task in natural language processing, especially in the context of social media where sarcasm is prevalent. This paper proposes a novel model called KnowleNet that incorporates prior knowledge and cross-modal semantic contrast for multimodal sarcasm detection. By leveraging the ConceptNet knowledge base and utilizing contrastive learning, the model achieves state-of-the-art performance on benchmark datasets.
INFORMATION FUSION
(2023)
Article
Computer Science, Artificial Intelligence
Deeksha Varshney, Asif Ekbal, Erik Cambria
Summary: This paper focuses on the neural-based interactive dialogue system that aims to engage and retain humans in long-lasting conversations. It proposes a new neural generative model that combines step-wise co-attention, self-attention-based transformer network, and an emotion classifier to control emotion and knowledge transfer during response generation. The results from quantitative, qualitative, and human evaluation show that the proposed models can generate natural and coherent sentences, capturing essential facts with significant improvement over emotional content.
KNOWLEDGE-BASED SYSTEMS
(2024)
Article
Computer Science, Artificial Intelligence
Jialun Wu, Kai He, Rui Mao, Chen Li, Erik Cambria
Summary: Predicting a patient's future health condition is a trending topic in the intelligent medical field. This paper proposes a knowledge-guided predictive framework called MEGACare, which leverages multi-faceted medical knowledge and multi-view learning to enhance clinical prediction accuracy.
INFORMATION FUSION
(2023)
Article
Computer Science, Artificial Intelligence
Kai He, Rui Mao, Yucheng Huang, Tieliang Gong, Chen Li, Erik Cambria
Summary: In this paper, a prompt-based contrastive learning method is proposed for few-shot NER tasks. The method leverages external knowledge to initialize semantic anchors and optimizes prompts and sentence embeddings with a proposed semantic-enhanced contrastive loss. The method outperforms traditional contrastive learning methods in few-shot scenarios and effectively addresses the issues in conventional methods.
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
(2023)