We study the ability of foundation models to learn representations for classification that are transferable to new, unseen classes. Recent results in the literature show that representations learned by a single classifier over many classes are competitive on few-shot learning problems with representations learned by special-purpose algorithms designed for such problems. We offer a theoretical explanation for this behavior based on the recently discovered phenomenon of class-feature-variability collapse, that is, that during the training of deep classification networks the feature embeddings of samples belonging to the same class tend to concentrate around their class means. More specifically, we show that the few-shot error of the learned feature map on new classes (defined as the classification error of the nearest class-center classifier using centers learned from a small number of random samples from each new class) is small in case of class-feature-variability collapse, under the assumption that the classes are selected independently from a fixed distribution. This suggests that foundation models can provide feature maps that are transferable to new downstream tasks, even with very few samples; to our knowledge, this is the first performance bound for transfer-learning that is non-vacuous in the few-shot setting. Keywords: Generalization bounds, foundation models, few-shot learning, transfer learning, neural collapse, class-features variability collapse.

Harriet Tubman didn't give many interviews in her lifetime, and when she did, they were generally conducted by one of her friends, Sarah Hopkins Bradford, a White children's book author in Upstate New York, where Tubman spent the last decades of her life. The result of those interviews were two biographies, published in 1869 and 1886. Though Bradford obviously admired Tubman, the books suffer from her sometimes patronizing attitude toward her subject, her use of racial slurs and her awkward attempts to re-create the speech patterns of a Black woman raised enslaved in Maryland. Some of the long "quotes" from Tubman were completely made up, and it shows. So I was curious to see what would happen recently when I had my own "interview" with Tubman -- using the online educator Khan Academy's new artificial intelligence learning tool Khanmigo, which enables users to have live chats with dozens of simulated historical figures like Abigail Adams, Genghis Khan, Montezuma and Winston Churchill. And if so, would it come off horribly, a 21st-century minstrelsy?

Presently, knowledge graph-based recommendation algorithms have garnered considerable attention among researchers. However, these algorithms solely consider knowledge graphs with single relationships and do not effectively model exercise-rich features, such as exercise representativeness and informativeness. Consequently, this paper proposes a framework, namely the Knowledge-Graph-Exercise Representativeness and Informativeness Framework, to address these two issues. The framework consists of four intricate components and a novel cognitive diagnosis model called the Neural Attentive cognitive diagnosis model. These components encompass the informativeness component, exercise representation component, knowledge importance component, and exercise representativeness component. The informativeness component evaluates the informational value of each question and identifies the candidate question set that exhibits the highest exercise informativeness. Furthermore, the skill embeddings are employed as input for the knowledge importance component. This component transforms a one-dimensional knowledge graph into a multi-dimensional one through four class relations and calculates skill importance weights based on novelty and popularity. Subsequently, the exercise representativeness component incorporates exercise weight knowledge coverage to select questions from the candidate question set for the tested question set. Lastly, the cognitive diagnosis model leverages exercise representation and skill importance weights to predict student performance on the test set and estimate their knowledge state. To evaluate the effectiveness of our selection strategy, extensive experiments were conducted on two publicly available educational datasets. The experimental results demonstrate that our framework can recommend appropriate exercises to students, leading to improved student performance.

The more new features that are being added to smartphones, the harder it becomes for users to find them. This is because the feature names are usually short, and there are just too many to remember. In such a case, the users may want to ask contextual queries that describe the features they are looking for, but the standard term frequency-based search cannot process them. This paper presents a novel retrieval system for mobile features that accepts intuitive and contextual search queries. We trained a relevance model via contrastive learning from a pre-trained language model to perceive the contextual relevance between query embeddings and indexed mobile features. Also, to make it run efficiently on-device using minimal resources, we applied knowledge distillation to compress the model without degrading much performance. To verify the feasibility of our method, we collected test queries and conducted comparative experiments with the currently deployed search baselines. The results show that our system outperforms the others on contextual sentence queries and even on usual keyword-based queries.

We propose an efficient modeling framework for cross-lingual named entity recognition in semi-structured text data. Our approach relies on both knowledge distillation and consistency training. The modeling framework leverages knowledge from a large language model (XLMRoBERTa) pre-trained on the source language, with a student-teacher relationship (knowledge distillation). The student model incorporates unsupervised consistency training (with KL divergence loss) on the low-resource target language. We employ two independent datasets of SMSs in English and Arabic, each carrying semi-structured banking transaction information, and focus on exhibiting the transfer of knowledge from English to Arabic. With access to only 30 labeled samples, our model can generalize the recognition of merchants, amounts, and other fields from English to Arabic. We show that our modeling approach, while efficient, performs best overall when compared to state-of-the-art approaches like DistilBERT pre-trained on the target language or a supervised model directly trained on labeled data in the target language. Our experiments show that it is enough to learn to recognize entities in English to reach reasonable performance in a low-resource language in the presence of a few labeled samples of semi-structured data. The proposed framework has implications for developing multi-lingual applications, especially in geographies where digital endeavors rely on both English and one or more low-resource language(s), sometimes mixed with English or employed singly.

"Mens sana in corpore sano" (Juvenal, 100-127 AD) Mental training has long been part of human culture, appearing in several different forms ranging from meditation to particular games or cognitive exercises aimed at various purposes. The past decades, however, have seen it losing its cardinal role in the well-roundedness of an individual and becoming more of a side hustle, confined to particular hobbies or to specific techniques needed for mental-health purposes. By contrast, recent years have seen an exponential investment in and development of artificial intelligence and machine learning technologies, which seem to be successfully tackling increasingly difficult tasks and problems. This work, then, aims to combine the two fields together by presenting a practical implementation of machine learning to the particular form of mental training that is the art of memory, taken in its competitive version called "Memory Sports". Such a fusion, on the one hand, strives to raise awareness about both realms, while on the other it seeks to encourage research in this mixed field as a way to, ultimately, drive forward the development of this seemingly underestimated sport. After first introducing the topic of mental training and its particular branch of Memory Sports, in the first chapter, the machine learning involved in the project is explained in the second chapter. The third chapter, then, presents two practical implementations of machine learning in Memory Sports, the results of which are discussed in the final chapter, together with several potential directions for future research. Ultimately, as well as stimulating interest and inspiration regarding the two fields involved in this research and exploring their points of contact, the aim here is also to investigate potential developments of human-machine collaborations, which are likely to be the focus of the next advances in technology and society overall. Starting to think in this view can help better prepare for the abrupt changes that might come and even be part of them, so as to drive their aim and scope toward a more responsible, and thus better, outcome.

Raven's Progressive Matrix (RPM) is a classic test to realize such ability in machine intelligence by selecting from candidates. Recent studies suggest that solving RPM in an answer-generation way boosts a more in-depth understanding of rules. However, existing generative solvers cannot discover the global concept-changing rules without auxiliary supervision (e.g., rule annotations and distractors in candidate sets). To this end, we propose a deep latent variable model for Concept-changing Rule ABstraction (CRAB) by learning interpretable concepts and parsing concept-changing rules in the latent space. With the iterative learning process, CRAB can automatically abstract global rules shared on the dataset on each concept and form the learnable prior knowledge of global rules. CRAB outperforms the baselines trained without auxiliary supervision in the arbitrary-position answer generation task and achieves comparable and even higher accuracy than the compared models trained with auxiliary supervision. Finally, we conduct experiments to illustrate the interpretability of CRAB in concept learning, answer selection, and global rule abstraction.

Graph neural networks have become the standard approach for dealing with learning problems on graphs. Among the different variants of graph neural networks, graph attention networks (GATs) have been applied with great success to different tasks. In the GAT model, each node assigns an importance score to its neighbors using an attention mechanism. However, similar to other graph neural networks, GATs aggregate messages from nodes that belong to different classes, and therefore produce node representations that are not well separated with respect to the different classes, which might hurt their performance. In this work, to alleviate this problem, we propose a new technique that can be incorporated into any graph attention model to encourage higher attention scores between nodes that share the same class label. We evaluate the proposed method on several node classification datasets demonstrating increased performance over standard baseline models.

The task of Argument Mining, that is extracting argumentative sentences for a specific topic from large document sources, is an inherently difficult task for machine learning models and humans alike, as large Argument Mining datasets are rare and recognition of argumentative sentences requires expert knowledge. The task becomes even more difficult if it also involves stance detection of retrieved arguments. Given the cost and complexity of creating suitably large Argument Mining datasets, we ask whether it is necessary for acceptable performance to have datasets growing in size. Our findings show that, when using carefully composed training samples and a model pretrained on related tasks, we can reach 95% of the maximum performance while reducing the training sample size by at least 85%. This gain is consistent across three Argument Mining tasks on three different datasets. We also publish a new dataset for future benchmarking.

Education strikes at the heart of what makes us human. It drives the intellectual capacity and prosperity of nations. It has developed the minds that took us to the moon and eradicated previously incurable diseases. And the special status of education is why generative AI tools such as ChatGPT are likely to profoundly disrupt this sector. This isn't a reflection of their intelligence, but of our failure to build education systems that nurture and value our unique human intelligence.