Conversational teachable agents offer a promising platform to support learning, both in the classroom and in remote settings. In this context, the agent takes the role of the novice, while the student takes on the role of teacher. This framing is significant for its ability to elicit the Prot\'eg\'e effect in the student-teacher, a pedagogical phenomenon known to increase engagement in the teaching task, and also improve cognitive outcomes. In prior work, teachable agents often take a passive role in the learning interaction, and there are few studies in which the agent and student engage in natural language dialogue during the teaching task. This work investigates the effect of teaching modality when interacting with a virtual agent, via the web-based teaching platform, the Curiosity Notebook. A method of teaching the agent by selecting sentences from source material is compared to a method paraphrasing the source material and typing text input to teach. A user study has been conducted to measure the effect teaching modality on the learning outcomes and engagement of the participants. The results indicate that teaching via paraphrasing and text input has a positive effect on learning outcomes for the material covered, and also on aspects of affective engagement. Furthermore, increased paraphrasing effort, as measured by the similarity between the source material and the material the teacher conveyed to the robot, improves learning outcomes for participants.
Decision forest algorithms model data by learning a binary tree structure recursively where every node splits the feature space into two regions, sending examples into the left or right branches. This ``decision'' is the result of the evaluation of a condition. For example, a node may split input data by applying a threshold to a numerical feature value. Such decisions are learned using (often greedy) algorithms that attempt to optimize a local loss function. Crucially, whether an algorithm exists to find and evaluate splits for a feature type (e.g., text) determines whether a decision forest algorithm can model that feature type at all. In this work, we set out to devise such an algorithm for textual features, thereby equipping decision forests with the ability to directly model text without the need for feature transformation. Our algorithm is efficient during training and the resulting splits are fast to evaluate with our extension of the QuickScorer inference algorithm. Experiments on benchmark text classification datasets demonstrate the utility and effectiveness of our proposal.
Training a multi-speaker Text-to-Speech (TTS) model from scratch is computationally expensive and adding new speakers to the dataset requires the model to be re-trained. The naive solution of sequential fine-tuning of a model for new speakers can cause the model to have poor performance on older speakers. This phenomenon is known as catastrophic forgetting. In this paper, we look at TTS modeling from a continual learning perspective where the goal is to add new speakers without forgetting previous speakers. Therefore, we first propose an experimental setup and show that serial fine-tuning for new speakers can result in the forgetting of the previous speakers. Then we exploit two well-known techniques for continual learning namely experience replay and weight regularization and we reveal how one can mitigate the effect of degradation in speech synthesis diversity in sequential training of new speakers using these methods. Finally, we present a simple extension to improve the results in extreme setups.
Self-supervised pre-training methods have brought remarkable breakthroughs in the understanding of text, image, and speech. Recent developments in genomics has also adopted these pre-training methods for genome understanding. However, they focus only on understanding haploid sequences, which hinders their applicability towards understanding genetic variations, also known as single nucleotide polymorphisms (SNPs), which is crucial for genome-wide association study. In this paper, we introduce SNP2Vec, a scalable self-supervised pre-training approach for understanding SNP. We apply SNP2Vec to perform long-sequence genomics modeling, and we evaluate the effectiveness of our approach on predicting Alzheimer's disease risk in a Chinese cohort. Our approach significantly outperforms existing polygenic risk score methods and all other baselines, including the model that is trained entirely with haploid sequences. We release our code and dataset on https://github.com/HLTCHKUST/snp2vec.
It has been widely known that CAM (Class Activation Map) usually only activates discriminative object regions and falsely includes lots of object-related backgrounds. As only a fixed set of image-level object labels are available to the WSSS (weakly supervised semantic segmentation) model, it could be very difficult to suppress those diverse background regions consisting of open set objects. In this paper, we propose a novel Cross Language Image Matching (CLIMS) framework, based on the recently introduced Contrastive Language-Image Pre-training (CLIP) model, for WSSS. The core idea of our framework is to introduce natural language supervision to activate more complete object regions and suppress closely-related open background regions. In particular, we design object, background region and text label matching losses to guide the model to excite more reasonable object regions for CAM of each category. In addition, we design a co-occurring background suppression loss to prevent the model from activating closely-related background regions, with a predefined set of class-related background text descriptions. These designs enable the proposed CLIMS to generate a more complete and compact activation map for the target objects. Extensive experiments on PASCAL VOC2012 dataset show that our CLIMS significantly outperforms the previous state-of-the-art methods. Code will be available.
Although adapting pre-trained language models with few examples has shown promising performance on text classification, there is a lack of understanding of where the performance gain comes from. In this work, we propose to answer this question by interpreting the adaptation behavior using post-hoc explanations from model predictions. By modeling feature statistics of explanations, we discover that (1) without fine-tuning, pre-trained models (e.g. BERT and RoBERTa) show strong prediction bias across labels; (2) although few-shot fine-tuning can mitigate the prediction bias and demonstrate promising prediction performance, our analysis shows models gain performance improvement by capturing non-task-related features (e.g. stop words) or shallow data patterns (e.g. lexical overlaps). These observations alert that pursuing model performance with fewer examples may incur pathological prediction behavior, which requires further sanity check on model predictions and careful design in model evaluations in few-shot fine-tuning.
Scene text detection is still a challenging task, as there may be extremely small or low-resolution strokes, and close or arbitrary-shaped texts. In this paper, StrokeNet is proposed to effectively detect the texts by capturing the fine-grained strokes, and infer structural relations between the hierarchical representation in the graph. Different from existing approaches that represent the text area by a series of points or rectangular boxes, we directly localize strokes of each text instance through Stroke Assisted Prediction Network (SAPN). Besides, Hierarchical Relation Graph Network (HRGN) is adopted to perform relational reasoning and predict the likelihood of linkages, effectively splitting the close text instances and grouping node classification results into arbitrary-shaped text region. We introduce a novel dataset with stroke-level annotations, namely SynthStroke, for offline pre-training of our model. Experiments on wide-ranging benchmarks verify the State-of-the-Art performance of our method. Our dataset and code will be available.
Past works that investigate out-of-domain performance of QA systems have mainly focused on general domains (e.g. news domain, wikipedia domain), underestimating the importance of subdomains defined by the internal characteristics of QA datasets. In this paper, we extend the scope of "out-of-domain" by splitting QA examples into different subdomains according to their several internal characteristics including question type, text length, answer position. We then examine the performance of QA systems trained on the data from different subdomains. Experimental results show that the performance of QA systems can be significantly reduced when the train data and test data come from different subdomains. These results question the generalizability of current QA systems in multiple subdomains, suggesting the need to combat the bias introduced by the internal characteristics of QA datasets.
Few-shot named entity recognition (NER) systems aim at recognizing novel-class named entities based on only a few labeled examples. In this paper, we present a decomposed meta-learning approach which addresses the problem of few-shot NER by sequentially tackling few-shot span detection and few-shot entity typing using meta-learning. In particular, we take the few-shot span detection as a sequence labeling problem and train the span detector by introducing the model-agnostic meta-learning (MAML) algorithm to find a good model parameter initialization that could fast adapt to new entity classes. For few-shot entity typing, we propose MAML-ProtoNet, i.e., MAML-enhanced prototypical networks to find a good embedding space that can better distinguish text span representations from different entity classes. Extensive experiments on various benchmarks show that our approach achieves superior performance over prior methods.
Text-to-speech (TTS) offers the opportunity to compensate for a hearing loss at the source rather than correcting for it at the receiving end. This removes limitations such as time constraints for algorithms that amplify a sound individually and can lead to higher speech quality for hearing-impaired listeners. We propose an algorithm that restores loudness to normal perception at a high resolution in time, frequency and level, and embed it in a TTS system that uses Tacotron2 and WaveGlow to produce individually amplified speech. Subjective evaluations of speech quality showed that the proposed algorithm led to high-quality audio. Mean opinion scores were predicted well by the STOI metric. Transfer learning led to a quick adaption of the produced spectra from original speech to individually amplified speech and gives us a way to train an individual TTS system efficiently.