Knowledge distillation aims to find a smaller model (i.e., student) that can perform at the level of a larger model (i.e., teacher). While the student model is highly beneficial in resource-constrained environments, finding the optimal student remains challenging due to the extensive search required through potential architectures and hyperparameters. To address this, we introduce a novel framework that integrates a caching mechanism and proximity analysis into Reinforcement Learning (RL) for Neural Architecture Search (NAS) and Hyperparameter Optimization (HPO) in knowledge distillation. This approach improves computational efficiency by avoiding redundant evaluations and estimating the performance of similar configurations. Our results, benchmarked against foundational and modern evolutionary search methods, demonstrate that the proposed framework can reduce full training evaluations by over 75% relative to a standard RL search, offering a robust advantage in computationally or operationally constrained environments.

Affective tasks, including sentiment analysis, emotion classification, and sarcasm detection have drawn a lot of attention in recent years due to a broad range of useful applications in various domains. The main goal of affect detection tasks is to recognizestatessuch as mood, sentiment, and emotions from textual data (e.g., news articles or product reviews). Despite the importance of utilizingpreprocessingsteps in different stages (i.e., word representation learning and building a classification model) of affect detection tasks, this topic has not been studied well. To that end, we explore whether applying various preprocessing methods (stemming, lemmatization, stopword removal, punctuation removal and so on) and their combinations in different stages of the affect detection pipeline can improve the model performance. The are many preprocessing approaches that can be utilized in affect detection tasks. However, their influence on the final performance depends on the type of preprocessing and the stages that they are applied. Moreover, the preprocessing impacts vary across different affective tasks. Our analysis provides thorough insights into how preprocessing steps can be applied in building an effect detection pipeline and their respective influence on performance.

Rapidly spreading diseases pose a significant threat, leading to substantial loss of life and economic devastation, as seen in the global COVID-19 outbreaks. Developing disease prediction models is crucial for preemptive pandemic control and minimizing their impact. As internet accessibility grows through computers and mobile devices, social media platforms provide a direct conduit to disseminate vital health information to the public. Unlike traditional methods that rely on bureaucratic channels, these platforms offer accurate and timely information distribution. We propose a framework that employs ontology to identify these symptoms and gather relevant tweets. Subsequently, the XGBoost-BiLSTM hybrid model harnesses this data to predict the count of infected cases. This hybrid model capitalizes on XGBoost’s prowess in handling limited dataset sizes, a prevalent challenge during outbreaks with insufficient time series data. Moreover, it enriches data for BiLSTM, amplifying its efficacy in predicting and monitoring outbreaks. To construct our dataset, we extracted tweets discussing symptoms from six distinct infectious disease outbreaks (Ebola, Zika, MERS, H1N1, Chikungunya, COVID-19) spanning from 2012 to 2021. Our results demonstrate that the proposed hybrid model outperforms nine cutting-edge and baseline models. This advancement can significantly assist health authorities in minimizing fatalities and preparing preemptively for potential outbreaks.

The advent of transformers and the subsequent development of Large Language Models (LLMs) based on these technologies has revolutionized the field of Natural Language Processing (NLP). These models are able to understand and generate coherent natural language and hold conversations with humans continuously. Meanwhile, ChatGPT has become famous among many LLMs for its general-purpose characteristics and versatility. With that in mind, we investigate the capabilities of ChatGPT, which is very successful in many downstream NLP tasks on the task of Question Generation (QG). In particular, our experiments show that appropriate context through our designed prompts makes ChatGPT an appropriate tool for accurately performing the QG task. We compare ChatGPT’s question generation results with the state-of-the-art models, particularly on the SQuAD and car manual datasets. The results show that ChatGPT is able to compete with or even outperform some of the baseline models. Furthermore, we illustrate that we may improve ChatGPT through additional fine-tuning of the prompts. Finally, we also investigate the use of ChatGPT to evaluate QG models. While the use of ChatGPT for such purposes is still in its early stages, our results demonstrate that ChatGPT can potentially be a strong QG accuracy evaluator comparable to human evaluators.

Automatic grading of computer programs has a great impact on both computer science education and the software industry as it saves human evaluators a tremendous amount of time required for assessing programs. However, to date, this problem lacks extensive research from the machine learning/deep learning perspective. Currently, the traditional auto-grading systems are mostly based on test-case execution results. However, these approaches lack insight into the syntax and semantics of the codes, and therefore, are far from human-level evaluation. In this study, we leverage the power of language models pre-trained on programming languages. We introduce two simple deep architectures and show that they consistently outperform the shallow models built upon extensive feature engineering approaches by a high margin. We also develop an incremental transductive learning algorithm that only requires a single reference solution to a problem and takes advantage of the correct implementations in the set of programs to be evaluated. Furthermore, our human evaluation results show that the proposed approaches provide partial marks having a strong correlation with marks given by human graders. We prepare and share a dataset of C++ and Python programs for future research (Code and data are available at https://github.com/peter-nagy1/Deep-Grader).

Automatic generation of questions from text has gained increasing attention due to its useful applications.We propose a novel question generation method that combines the benefits of rule-based and neural sequence-to-sequence (Seq2Seq) models.The proposed method can automatically generate multiple questions from an input sentence covering different views of the sentence as in rule-based methods, while more complicated "rules" can be learned via the Seq2Seq model.The method utilizes semantic role labeling to convert training examples into their semantic representations, and then trains a Seq2Seq model over the semantic representations.Our extensive experiments on three realworld data sets show that the proposed method significantly improves the state-of-the-art neural question generation approaches.

The Proximal Policy Optimization (PPO) is a policy gradient approach providing state-of-the-art performance in many domains through the “surrogate” objective function using stochastic gradient ascent. While PPO is an appealing approach in reinforcement learning, it does not consider the importance of states (a frequently seen state in a successful trajectory) in policy/value function updates. In this work, we introduce Preferential Proximal Policy Optimization (P3O) which incorporates the importance of these states into parameter updates. First, we determine the importance of each state based on the variance of the action probabilities given a particular state multiplied by the value function, normalized and smoothed using the Exponentially Weighted Moving Average. Then, we incorporate the state's importance in the surrogate objective function. That is, we redefine value and advantage estimation objectives functions in the PPO approach. Unlike other related approaches, we select the importance of states automatically which can be used for any algorithm utilizing a value function. Empirical evaluations across six Atari environments demonstrate that our approach significantly outperforms the baseline (vanilla PPO) across different tested environments, highlighting the value of our proposed method in learning complex environments.

A common approach in reinforcement learning (RL) is to give the agent a static reward for successfully completing the task or punishing it for failing. However, this approach leads to a behaviour similar to fear in combat scenarios. It learns a sub-optimal policy improving over time while retaining elements of cowardice in updating the policy. Cowardice can be avoided by removing static rewards given to the agent at the terminal state, but this lack of reward can negatively affect performance. This paper presents a novel approach to solve these issues by decaying this reward or punishment based on the agent’s performance at the terminal state and evaluates the proposed method across three separate games of varying levels of complexity—The Legend of Zelda, Megaman X, and M.U.G.E.N. All three games are based on combat scenarios where the goal is to defeat the opponent by reducing its health to zero. In all environments, the agents receiving decayed reward and punishment are more stable when training, achieve higher win rates, and require fewer actions per game than their statically rewarded counterparts.

There is a growing need to find meaningful teams in expert networks such as DBLP and GitHub. However, existing team formation methods, such as those based on shortest paths between experts, may generate weakly-connected teams. We demonstrate RW-Team, a robust team formation framework based on a random walk with restart (RWR). We introduce a greedy algorithm to reduce the search space, and we use a Monte Carlo approximation of RWR to improve performance. To handle large graphs, we implement RW-Team in Apache Spark. The proposed demonstration will allow participants to form teams of researchers having various skill sets and explore connections among team members using several graph visualization techniques.

Subscription-based online newspapers usually offer non-subscribed users a certain number of free articles in a period of time, and then directs them to a page (called paywall) asking for subscription. This approach (also known as metered or fixed paywall) does not consider the user's reading history nor the articles that the user may read in the future, and consequently, it may disengage many potential subscribers. To that end, we propose adaptive paywall mechanisms to make optimal paywall decisions (i.e., showing the article or the paywall) by balancing the benefit of showing the article against that of presenting the paywall. We define the notions of utility and cost which are used to define an objective function for the optimal paywall decision problem. We propose the Lookahead policy (LAP) and QPaywall policy (QP) as two data-driven approaches to solve the adaptive paywall problem. While the LAP method makes paywall decisions on the fly by simulating trajectories of article requests using Monte Carlo sampling, the QP approach is based on reinforcement learning and learns a neural network-based action-value (Q) function for this purpose. We compare advantages of the proposed approaches and discuss the practical considerations of using them in a real environment. Empirical studies on a real dataset from a major newspaper in Canada show that the proposed methods outperform several baseline approaches in terms of various business objectives.

Automatic sarcasm detection from text is an important classification task that can help identify the actual sentiment in user-generated data, such as reviews or tweets.Despite its usefulness, sarcasm detection remains a challenging task, due to a lack of any vocal intonation or facial gestures in textual data.To date, most of the approaches to addressing the problem have relied on hand-crafted affect features, or pre-trained models of non-contextual word embeddings, such as Word2vec.However, these models inherit limitations that render them inadequate for the task of sarcasm detection.In this paper, we propose two novel deep neural network models for sarcasm detection, namely ACE 1 and ACE 2. Given as input a text passage, the models predict whether it is sarcastic (or not).Our models extend the architecture of BERT by incorporating both affective and contextual features.To the best of our knowledge, this is the first attempt to directly alter BERT's architecture and train it from scratch to build a sarcasm classifier.Extensive experiments on different datasets demonstrate that the proposed models outperform state-of-the-art models for sarcasm detection with significant margins.

Many online news agencies utilize the paywall mechanism to increase reader subscriptions. This method offers a non-subscribed reader a fixed number of free articles in a period of time (e.g., a month), and then directs the user to the subscription page for further reading. We argue that there is no direct relationship between the number of paywalls presented to readers and the number of subscriptions, and that this artificial barrier, if not used well, may disengage potential subscribers and thus may not well serve its purpose of increasing revenue. Moreover, the current paywall mechanism neither considers the user browsing history nor the potential articles which the user may visit in the future. Thus, it treats all readers equally and does not consider the potential of a reader in becoming a subscriber. In this paper, we propose an adaptive paywall mechanism to balance the benefit of showing an article against that of displaying the paywall (i.e., terminating the session). We first define the notion of cost and utility that are used to define an objective function for optimal paywall decision making. Then, we model the problem as a stochastic sequential decision process. Finally, we propose an efficient policy function for paywall decision making. The experimental results on a real dataset from a major newspaper in Canada show that the proposed model outperforms the traditional paywall mechanism as well as the other baselines.

User acquisition is one of the most challenging problems for online news providers. In fact, due to availability of different news media, users have a lot of choices in selecting the news source. To date, most of digital news portals have tried to approach the solution indirectly by targeting the user satisfaction through the recommendation systems. In contrast, we address the problem directly by identifying valuable visitors who are likely potential subscribers in the future. First, we suggest that the decision for subscription is not a sudden, instantaneous action, but is the informed decision based on positive experience with digital medium. As such, we propose effective engagement measures and show that they are effective in building the predictive model for subscription. We design a model that not only predicts the potential subscribers but also answers queries about the subscription occurrence time. The proposed model can be used to predict the subscription time and recommend accurately the “potential users” to the current marketing campaign. We evaluate the proposed model using a real dataset from The Globe and Mail which is a major newspaper in Canada. The experimental results show that the proposed model outperforms the traditional state-of-the-art approaches significantly.

Sequential pattern mining has been used in bioinformatics to discover frequent gene regulation sequential patterns based on time course microarray datasets. While mining frequent sequences are important in biological studies for disease treatment, to date, most of the approaches do not consider the importance of the genes with respect to a disease being studied when identifying gene regulation sequential patterns. In addition, they focus on the more general up/down effects of genes in a microarray dataset and do not take into account the various degrees of expression during the mining process. As a result, the current techniques return too many sequences which may not be informative enough for biologists to explore relationships between the disease and underlying causes encoded in gene regulation sequences. In this paper, we propose a utility model by considering both the importance of genes with respect to a disease and their degrees of expression levels under a biological investigation. Then, we design a new method, called TU-SEQ, for identifying top-k high utility gene regulation sequential patterns from a time-course microarray dataset. The evaluation results show that our approach can effectively and efficiently discover key patterns representing meaningful gene regulation sequential patterns in a time course microarray dataset.

Forced migration is increasingly becoming a global issue of concern. In this paper, we present an effective model of targeted event detection, as an essential step towards the forced migration detection problem. To date, most of the the approaches deal with the event detection in a general setting with the main objective of detecting the presence or onset of an event. However, we focus on analyzing the magnitude of a given event from a collection of text documents such as news articles from multiple sources. We use violence as an illustration as it is one of the most critical factors of forced migration. The recent advancements in semantic similarity measures are adopted to obtain relevant violence scores for each word in the vocabulary of news articles in an unsupervised manner. The resulting scores are then used to compute the average daily violence scores over a period of three months. Evaluation of the proposed model against a manually annotated data set yields a Pearson's correlation of 0.8. We also include a case study exploring the relationship between violence and key events.