The performance of Vision–Language Transformers drops sharply when an input modality (e.g., image) is missing, because the model is forced to make predictions using incomplete information. Existing missing-aware prompt methods help reduce this degradation, but they still rely on conventional prediction heads (e.g., a Fully-Connected layer) that compute class scores in the same way regardless of which modality is present or absent. We introduce Decoupled Prototype Learning (DPL), a new prediction head architecture that explicitly adjusts its decision process to the observed input modalities. For each class, DPL selects a set of prototypes specific to the current missing-modality cases (image-missing, text-missing, or mixed-missing). Each prototype is then decomposed into image-specific and text-specific components, enabling the head to make decisions that depend on the information actually present. This adaptive design allows DPL to handle inputs with missing modalities more effectively while remaining fully compatible with existing prompt-based frameworks. Extensive experiments on MM-IMDb, UPMC Food-101, and Hateful Memes demonstrate that DPL outperforms state-of-the-art approaches across all widely used multimodal image–text datasets and various missing cases.

Active learning, which lets the model select the most informative data within a limited annotation budget for annotation, often achieves comparable performance with fewer labelled samples. However, despite its potential benefits, active learning has not been extensively explored in Named Entity Recognition (NER), as NER is considered more challenging than traditional tasks commonly addressed in the active learning literature. In this paper, we propose a novel deep active learning framework called Deep AUC Maximisation-based Active Learning (DAMAL), which learns uncertainty-aware representations for unlabeled data. Notably, DAMAL, due to its foundation in AUC maximisation, excels in addressing label sparsity, making it a well-suited approach for NER tasks. Experimental results across several NER tasks highlight the advantages of the proposed framework over existing active learning methods for NER.

Automated explanatory feedback systems play a crucial role in facilitating learning for a large cohort of learners by offering feedback that incorporates explanations, significantly enhancing the learning process. However, delivering such explanatory feedback in real-time poses challenges, particularly when high classification accuracy for domain-specific, nuanced responses is essential. Our study leverages the capabilities of large language models, specifically Generative Pre-Trained Transformers (GPT), to explore a sequence labeling approach focused on identifying components of desired and less desired praise for providing explanatory feedback within a tutor training dataset. Our aim is to equip tutors with actionable, explanatory feedback during online training lessons. To investigate the potential of GPT models for providing the explanatory feedback, we employed two commonly-used approaches: prompting and fine-tuning. To quantify the quality of highlighted praise components identified by GPT models, we introduced a Modified Intersection over Union (M-IoU) score. Our findings demonstrate that: (1) the M-IoU score effectively correlates with human judgment in evaluating sequence quality; (2) using two-shot prompting on GPT-3.5 resulted in decent performance in recognizing effort-based (M-IoU of 0.46) and outcome-based praise (M-IoU of 0.68); and (3) our optimally fine-tuned GPT-3.5 model achieved M-IoU scores of 0.64 for effort-based praise and 0.84 for outcome-based praise, aligning with the satisfaction levels evaluated by human coders. Our results show promise for using GPT models to provide feedback that focuses on specific elements in their open-ended responses that are desirable or could use improvement.

The effectiveness of active learning largely depends on the sampling efficiency of the acquisition function. Expected Loss Reduction (ELR) focuses on a Bayesian estimate of the reduction in classification error, and more general costs fit in the same framework. We propose Bayesian Estimate of Mean Proper Scores (BEMPS) to estimate the increase in strictly proper scores such as log probability or negative mean square error within this framework. We also prove convergence results for this general class of costs. To facilitate better experimentation with the new acquisition functions, we develop a complementary batch AL algorithm that encourages diversity in the vector of expected changes in scores for unlabeled data. To allow high-performance classifiers, we combine deep ensembles, and dynamic validation set construction on pretrained models, and further speed up the ensemble process with the idea of Monte Carlo Dropout. Extensive experiments on both texts and images show that the use of mean square error and log probability with BEMPS yields robust acquisition functions and well-calibrated classifiers, and consistently outperforms the others tested. The advantages of BEMPS over the others are further supported by a set of qualitative analyses, where we visualise their sampling behaviour using data maps and t-SNE plots.

Automating the classification of instructional strategies from a large-scale online tutorial dialogue corpus is indispensable to the design of dialogue-based intelligent tutoring systems. Despite many existing studies employing supervised machine learning (ML) models to automate the classification process, they concluded that building a well-performed ML model is nontrivial since the sample size is commonly limited for model training. Based on this fact, we posited that the model performance can be further optimized by the design of input features and the selection of informative instances for model training. By reviewing the existing dialogue-related research, we found that contextual information (i.e., the content of preceding utterances) was a potential feature in many text classification tasks but underexplored in the classification of an instructional strategy. In addition, training a well-performed ML model (e.g., deep neural network) to recognize instructional strategies usually requires a large amount of manually annotated data, which are labor intensive. To alleviate the demand of manual annotation, researchers proposed the use of statistical active learning methods to select informative instances, but this method was rarely used for recognizing the instructional strategy in online tutoring dialogue. Therefore, our study aimed to investigate the improvement of automating the classification of dialogue acts—a popular approach to the detection of instructional strategies—from two perspectives. First, we explored whether and to what extent the incorporation of contextual information can boost a model's prediction performance. Then, we investigated the extent to which the recent active learning methods can alleviate the labor-intensive issues in training the ML model for recognizing the instructional strategies. Our study showed that: 1) the ML models trained on the features that included the contextual information achieved better performance than that of the models excluding it; 2) the effectiveness of the contextual information decayed after the ML model achieved an optimal performance; and 3) compared with the random baseline, active learning methods can select informative samples from the training dataset to train ML models, which can alleviate the labor-intensive issues.

Within the scope of natural language processing, the domain of multi-label text classification is uniquely challenging due to its expansive and uneven label distribution. The complexity deepens due to the demand for an extensive set of annotated data for training an advanced deep learning model, especially in specialized fields where the labeling task can be labor-intensive and often requires domain-specific knowledge. Addressing these challenges, our study introduces a novel deep active learning strategy, capitalizing on the Beta family of proper scoring rules within the Expected Loss Reduction framework. It computes the expected increase in scores using the Beta Scoring Rules, which are then transformed into sample vector representations. These vector representations guide the diverse selection of informative samples, directly linking this process to the model's expected proper score. Comprehensive evaluations across both synthetic and real datasets reveal our method's capability to often outperform established acquisition techniques in multi-label text classification, presenting encouraging outcomes across various architectural and dataset scenarios.

Large Language Models (LLMs) have recently gained significant attention due to their remarkable capabilities in performing diverse tasks across various domains. However, a thorough evaluation of these models is crucial before deploying them in real-world applications to ensure they produce reliable performance. Despite the well-established importance of evaluating LLMs in the community, the complexity of the evaluation process has led to varied evaluation setups, causing inconsistencies in findings and interpretations. To address this, we systematically review the primary challenges and limitations causing these inconsistencies and unreliable evaluations in various steps of LLM evaluation. Based on our critical review, we present our perspectives and recommendations to ensure LLM evaluations are reproducible, reliable, and robust.

Current work in named entity recognition (NER) uses either cross entropy (CE) or conditional random fields (CRF) as the objective/loss functions to optimize the underlying NER model. Both of these traditional objective functions for the NER problem generally produce adequate performance when the data distribution is balanced and there are sufficient annotated training examples. But since NER is inherently an imbalanced tagging problem, the model performance under the low-resource settings could suffer using these standard objective functions. Based on recent advances in area under the ROC curve (AUC) maximization, we propose to optimize the NER model by maximizing the AUC score. We give evidence that by simply combining two binary-classifiers that maximize the AUC score, significant performance improvement over traditional loss functions is achieved under low-resource NER settings. We also conduct extensive experiments to demonstrate the advantages of our method under the low-resource and highly-imbalanced data distribution settings. To the best of our knowledge, this is the first work that brings AUC maximization to the NER setting. Furthermore, we show that our method is agnostic to different types of NER embeddings, models and domains. The code of this work is available at https://github.com/dngu0061/NER-AUC-2T.

Named entity recognition (NER), a sequence labelling/token classification task, has been traditionally considered a multi-class classification problem, the learning objective of which is to either optimise the multi-class cross entropy loss (CE) or train a conditional random field (CRF). However, these standard learning objectives, though scalable to large NER datasets and used in state-of-the-art work, largely ignore the problem of imbalanced label distributions that is inherent in all NER corpora. We show this leads to degraded performance in low-resource settings. While reformulating this standard multi-class labelling problem as a one-vs-all (OVA) learning problem, we propose to optimise the NER model with an AUC-based alternative loss function that is more capable of handling imbalanced datasets. As OVA often leads to a higher training time compared to the standard multi-class setting, we also develop two training strategies, one trains together the labels that share similar linguistic characteristics, and another employs a meta-learning approach to speed convergence. In order to motivate some of our experiments and better interpret the results, we also develop a Bayesian theory for what is the AUC function during learning. Experimental results under low-resource NER settings from benchmark corpora show that our methods can achieve consistently better performance compared with the learning objectives commonly used in NER. We also give evidence that our methods are robust and agnostic to the underlying NER embeddings, models, domains, and label distributions. The code to replicate this work will be released upon the publication of this paper.

Active learning, a widely adopted technique for enhancing machine learning models in text and image classification tasks with limited annotation resources, has received relatively little attention in the domain of Named Entity Recognition (NER). The challenge of data imbalance in NER has hindered the effectiveness of active learning, as sequence labellers lack sufficient learning signals. To address these challenges, this paper presents a novel reweighting-based active learning strategy that assigns dynamic smoothed weights to individual tokens. This adaptable strategy is compatible with various token-level acquisition functions and contributes to the development of robust active learners. Experimental results on multiple corpora demonstrate the substantial performance improvement achieved by incorporating our re-weighting strategy into existing acquisition functions, validating its practical efficacy.

Personalized Learning Squared (PLUS) is a comprehensive tutoring platform developed by Carnegie Mellon University, Carnegie Learning, and Stanford University designed to provide human and artificial intelligence (AI)-assisted tutoring. By combining human tutors, AI-powered math software, and cutting edge learning engineering, PLUS strives to double the math learning for 10,000 students by the year 2026. PLUS focuses on improving math learning for middle school students, particularly economically and racially diverse students. In this proposal summary, we plan to introduce PLUS Training, which aims to square the impact of tutoring. PLUS Training can empower tutors through customized tutor training and continuous support using research-driven instructional materials and leveraging AI to generate real-time tutor feedback. Empowering and strengthening tutor impact fuels our vision of addressing the opportunity gap by increasing learning opportunities among historically marginalized students.

Dialogue acts (DAs) can represent conversational actions of tutors or students that take place during tutoring dialogues. Automating the identification of DAs in tutoring dialogues is significant to the design of dialogue-based intelligent tutoring systems. Many prior studies employ machine learning models to classify DAs in tutoring dialogues and invest much effort to optimize the classification accuracy by using limited amounts of training data (i.e., low-resource data scenario). However, beyond the classification accuracy, the robustness of the classifier is also important, which can reflect the capability of the classifier on learning the patterns from different class distributions. We note that many prior studies on classifying educational DAs employ cross entropy (CE) loss to optimize DA classifiers on low-resource data with imbalanced DA distribution. The DA classifiers in these studies tend to prioritize accuracy on the majority class at the expense of the minority class which might not be robust to the data with imbalanced ratios of different DA classes. To optimize the robustness of classifiers on imbalanced class distributions, we propose to optimize the performance of the DA classifier by maximizing the area under the ROC curve (AUC) score (i.e., AUC maximization). Through extensive experiments, our study provides evidence that (i) by maximizing AUC in the training process, the DA classifier achieves significant performance improvement compared to the CE approach under low-resource data, and (ii) AUC maximization approaches can improve the robustness of the DA classifier under different class imbalance ratios.

Dialogue Acts (DAs) can be used to explain what expert tutors do and what students know during the tutoring process. Most empirical studies adopt the random sampling method to obtain sentence samples for manual annotation of DAs, which are then used to train DA classifiers. However, these studies have paid little attention to sample informativeness, which can reflect the information quantity of the selected samples and inform the extent to which a classifier can learn patterns. Notably, the informativeness level may vary among the samples and the classifier might only need a small amount of low informative samples to learn the patterns. Random sampling may overlook sample informativeness, which consumes human labelling costs and contributes less to training the classifiers. As an alternative, researchers suggest employing statistical sampling methods of Active Learning (AL) to identify the informative samples for training the classifiers. However, the use of AL methods in educational DA classification tasks is under-explored. In this paper, we examine the informativeness of annotated sentence samples. Then, the study investigates how the AL methods can select informative samples to support DA classifiers in the AL sampling process. The results reveal that most annotated sentences present low informativeness in the training dataset and the patterns of these sentences can be easily captured by the DA classifier. We also demonstrate how AL methods can reduce the cost of manual annotation in the AL sampling process.

Research demonstrates learners engaging in the process of producing explanations to support their reasoning, can have a positive impact on learning. However, providing learners real-time explanatory feedback often presents challenges related to classification accuracy, particularly in domain-specific environments, containing situationally complex and nuanced responses. We present two approaches for supplying tutors real-time feedback within an online lesson on how to give students effective praise. This work-in-progress demonstrates considerable accuracy in binary classification for corrective feedback of effective, or effort-based (F1 score = 0.811), and ineffective, or outcome-based (F1 score = 0.350), praise responses. More notably, we introduce progress towards an enhanced approach of providing explanatory feedback using large language model-facilitated named entity recognition, which can provide tutors feedback, not only while engaging in lessons, but can potentially suggest real-time tutor moves. Future work involves leveraging large language models for data augmentation to improve accuracy, while also developing an explanatory feedback interface.

Named entity recognition (NER), a task that identifies and categorizes named entities such as persons or organizations from text, is traditionally framed as a multi-class classification problem. However, this approach often overlooks the issues of imbalanced label distributions, particularly in low-resource settings, which is common in certain NER contexts, like biomedical NER (bioNER). To address these issues, we propose an innovative reformulation of the multi-class problem as a one-vs-all (OVA) learning problem and introduce a loss function based on the area under the receiver operating characteristic curve (AUC). To enhance the efficiency of our OVA-based approach, we propose two training strategies: one groups labels with similar linguistic characteristics, and another employs meta-learning. The superiority of our approach is confirmed by its performance, which surpasses traditional NER learning in varying NER settings.

To construct dialogue-based Intelligent Tutoring Systems (ITS) with sufficient pedagogical expertise, a trendy research method is to mine large-scale data collected by existing dialogue-based ITS or generated between human tutors and students to discover effective tutoring strategies. However, most of the existing research has mainly focused on the analysis of successful tutorial dialogue. We argue that, to better inform the design of dialogue-based ITS, it is also important to analyse unsuccessful tutorial dialogues and gain a better understanding of the reasons behind those failures. Therefore, our study aimed to identify effective tutoring strategies by mining a large-scale dataset of both successful and unsuccessful human–human online tutorial dialogues, and further used these tutoring strategies for predicting students' problem-solving performance. Specifically, the study adopted a widely-used educational dialogue act scheme to describe the action behind utterances made by a tutor/student in the broader context of a tutorial dialogue (e.g., asking/answering a question, providing hints). Frequent dialogue acts were identified and analysed by taking into account the prior progress that a student had made before the start of a tutorial session and the problem-solving performance the student achieved after the end of the session. Besides, we performed a sequence analysis on the inferred actions to identify prominent patterns that were closely related to students' problem-solving performance. These prominent patterns could shed light on the frequent strategies used by tutors. Lastly, we measured the power of these tutorial actions in predicting students' problem-solving performance by applying a well-established machine learning method, Gradient Tree Boosting (GTB). Through extensive analysis and evaluations, we identified a set of different action patterns that were pertinent to tutors and students across dialogues of different traits, e.g., students without prior progress in solving problems, compared to those with prior progress, were likely to receive more thought-provoking questions from their tutors. More importantly, we demonstrated that the actions taken by students and tutors during a tutorial process could not adequately predict student performance and should be considered together with other relevant factors (e.g., the informativeness of the utterances).

We study acquisition functions for active learning (AL) for text classification. The Expected Loss Reduction (ELR) method focuses on a Bayesian estimate of the reduction in classification error, recently updated with Mean Objective Cost of Uncertainty (MOCU). We convert the ELR framework to estimate the increase in (strictly proper) scores like log probability or negative mean square error, which we call Bayesian Estimate of Mean Proper Scores (BEMPS). We also prove convergence results borrowing techniques used with MOCU. In order to allow better experimentation with the new acquisition functions, we develop a complementary batch AL algorithm, which encourages diversity in the vector of expected changes in scores for unlabelled data. To allow high performance text classifiers, we combine ensembling and dynamic validation set construction on pretrained language models. Extensive experimental evaluation then explores how these different acquisition functions perform. The results show that the use of mean square error and log probability with BEMPS yields robust acquisition functions, which consistently outperform the others tested.

This appendix contains further details regarding the theory and experiments. Section A gives the proofs of Lemma 1 about the properties of scoring rules and Theorem 1 about the convergence of active learning with (strictly) proper scoring functions. Section B provides details of the datasets, baseline implementation and experiment setup. Section C and Section D includes a comprehensive set of experimental results, including learning curves and pairwise comparison matrices on the four datasets that are not included in the main paper.