The field of language models has made significant progress in recent years. Large models like GPT-4 [17] have shown initial signs of general intelligence [8], while smaller models have demonstrated good reasoning abilities by solving challenging coding and math problems [11, 15, 16]. In this paper, we focus on the ability of small language models to solve grade-school math problems. Unlike previous works that empirically push the accuracy of models on grade-school math benchmarks like GSM8K [9] and its augmentations (e.g., [16, 22]), we take a more principled approach. We aim to understand the following fundamental questions: 1. How do language models learn to solve grade-school level math problems? Do they just memorize templates, or do they learn reasoning skills similar to humans? Or do they discover new skills to solve the problems?

Recent advancements in deep convolutional neural networks have significantly improved the performance of saliency prediction. However, the manual configuration of the neural network architectures requires domain knowledge expertise and can still be time-consuming and error-prone. To solve this, we propose a new Neural Architecture Search (NAS) framework for saliency prediction with two contributions. Firstly, a supernet for saliency prediction is built with a weight-sharing network containing all candidate architectures, by integrating a dynamic convolution into the encoder-decoder in the supernet, termed SalNAS. Secondly, despite the fact that SalNAS is highly efficient (20.98 million parameters), it can suffer from the lack of generalization. To solve this, we propose a self-knowledge distillation approach, termed Self-KD, that trains the student SalNAS with the weighted average information between the ground truth and the prediction from the teacher model. The teacher model, while sharing the same architecture, contains the best-performing weights chosen by cross-validation. Self-KD can generalize well without the need to compute the gradient in the teacher model, enabling an efficient training system. By utilizing Self-KD, SalNAS outperforms other state-of-the-art saliency prediction models in most evaluation rubrics across seven benchmark datasets while being a lightweight model. The code will be available at https://github.com/chakkritte/SalNAS

Instruction tuning in multimodal large language models (MLLMs) aims to smoothly integrate a backbone LLM with a pre-trained feature encoder for downstream tasks. The major challenge is how to efficiently find the synergy through cooperative learning where LLMs adapt their reasoning abilities in downstream tasks while feature encoders adjust their encoding to provide more relevant modal information. In this paper, we analyze the MLLM instruction tuning from both theoretical and empirical perspectives, where we find unbalanced learning between the two components, i.e., the feature encoder and the LLM, can cause diminishing learning gradients that slow the model convergence and often lead to sub-optimal results due to insufficient learning. Inspired by our findings, we propose a measurement to quantitatively evaluate the learning balance, based on which we further design a dynamic learning scheduler that better coordinates the learning. In addition, we introduce an auxiliary loss regularization method to promote updating of the generation distribution of MLLMs considering the learning state of each model component, which potentially prevents each component from gradient diminishing and enables a more accurate estimation of the learning balance coefficient. We conduct experiments with multiple LLM backbones and feature encoders, where our techniques are model-agnostic and can be generically integrated with various MLLM backbones. Experiment results on multiple downstream tasks and modalities in vision and audio, demonstrate the proposed method's better efficiency and effectiveness in MLLM instruction tuning.

Digital Simulations (DS) provide safe environments where users interact with an agent through conversational prompts, providing engaging learning experiences that can be used to train teacher candidates in realistic classroom scenarios. These simulations usually include open-ended questions, allowing teacher candidates to express their thoughts but complicating an automatic response analysis. To address this issue, we have evaluated Large Language Models (LLMs) to identify characteristics (user behaviors) in the responses of DS for teacher education. We evaluated the performance of DeBERTaV3 and Llama 3, combined with zero-shot, few-shot, and fine-tuning. Our experiments discovered a significant variation in the LLMs' performance depending on the characteristic to identify. Additionally, we noted that DeBERTaV3 significantly reduced its performance when it had to identify new characteristics. In contrast, Llama 3 performed better than DeBERTaV3 in detecting new characteristics and showing more stable performance. Therefore, in DS where teacher educators need to introduce new characteristics because they change depending on the simulation or the educational objectives, it is more recommended to use Llama 3. These results can guide other researchers in introducing LLMs to provide the highly demanded automatic evaluations in DS.

A Convolutional Recurrent Neural Network (CRNN) is trained to reproduce the evolution of the spinodal decomposition process in three dimensions as described by the Cahn-Hilliard equation. A specialized, physics-inspired architecture is proven to provide close accordance between the predicted evolutions and the ground truth ones obtained via conventional integration schemes. The method can closely reproduce the evolution of microstructures not represented in the training set at a fraction of the computational costs. Extremely long-time extrapolation capabilities are achieved, up to reaching the theoretically expected equilibrium state of the system, despite the training set containing only relatively-short, initial phases of the evolution. Quantitative accordance with the decay rate of the Free energy is also demonstrated up to late coarsening stages, providing an example of a data-driven, physically consistent and high-accuracy Machine Learning method for the long timescale simulation of materials.

Since the public release of Chat Generative Pre-Trained Transformer (ChatGPT), extensive discourse has emerged concerning the potential advantages and challenges of integrating Generative Artificial Intelligence (GenAI) into education. In the realm of information systems, research on technology adoption is crucial for understanding the diverse factors influencing the uptake of specific technologies. Theoretical frameworks, refined and validated over decades, serve as guiding tools to elucidate the individual and organizational dynamics, obstacles, and perceptions surrounding technology adoption. However, while several models have been proposed, they often prioritize elucidating the factors that facilitate acceptance over those that impede it, typically focusing on the student perspective and leaving a gap in empirical evidence regarding educators viewpoints. Given the pivotal role educators play in higher education, this study aims to develop a theoretical model to empirically predict the barriers preventing educators from adopting GenAI in their classrooms. Acknowledging the lack of theoretical models tailored to identifying such barriers, our approach is grounded in the Innovation Resistance Theory (IRT) framework and augmented with constructs from the Technology-Organization-Environment (TOE) framework. This model is transformed into a measurement instrument employing a quantitative approach, complemented by a qualitative approach to enrich the analysis and uncover concerns related to GenAI adoption in the higher education domain.

The emergence of Large Language Models (LLMs) has significantly impacted the field of Natural Language Processing and has transformed conversational tasks across various domains because of their widespread integration in applications and public access. The discussion surrounding the application of LLMs in education has raised ethical concerns, particularly concerning plagiarism and policy compliance. Despite the prowess of LLMs in conversational tasks, the limitations of reliability and hallucinations exacerbate the need to guardrail conversations, motivating our investigation of RAG in computer science higher education. We developed Retrieval Augmented Generation (RAG) applications for the two tasks of virtual teaching assistants and teaching aids. In our study, we collected the ratings and opinions of faculty members in undergraduate and graduate computer science university courses at various levels, using our personalized RAG systems for each course. This study is the first to gather faculty feedback on the application of LLM-based RAG in education. The investigation revealed that while faculty members acknowledge the potential of RAG systems as virtual teaching assistants and teaching aids, certain barriers and features are suggested for their full-scale deployment. These findings contribute to the ongoing discussion on the integration of advanced language models in educational settings, highlighting the need for careful consideration of ethical implications and the development of appropriate safeguards to ensure responsible and effective implementation.

Using questions in written text is an effective strategy to enhance readability. However, what makes an active reading question good, what the linguistic role of these questions is, and what is their impact on human reading remains understudied. We introduce GuidingQ, a dataset of 10K in-text questions from textbooks and scientific articles. By analyzing the dataset, we present a comprehensive understanding of the use, distribution, and linguistic characteristics of these questions. Then, we explore various approaches to generate such questions using language models. Our results highlight the importance of capturing inter-question relationships and the challenge of question position identification in generating these questions. Finally, we conduct a human study to understand the implication of such questions on reading comprehension. We find that the generated questions are of high quality and are almost as effective as human-written questions in terms of improving readers' memorization and comprehension.

This research combines Knowledge Distillation (KD) and Mixture of Experts (MoE) to develop modular, efficient multilingual language models. Key objectives include evaluating adaptive versus fixed alpha methods in KD and comparing modular MoE architectures for handling multi-domain inputs and preventing catastrophic forgetting. KD compresses large language models (LLMs) into smaller, efficient models, while MoE enhances modularity with specialized tasks. Experiments showed similar performance for both KD methods, with marginal improvements from adaptive alpha. A combined loss approach provided more stable learning. The router, trained to classify input sequences into English, French, German, or Python, achieved 99.95% precision, recall, and F1 score, with Logistic Regression being the most effective classifier. Evaluations of modular MoE architectures revealed that Pre-trained Language Experts (PLE) and Joint Expert Embedding Training (JEET) performed similarly, while the MoE with Common Expert (MoE-CE) setup showed slightly lower performance. Including a common expert in MoE-CE improved its performance. Studies on catastrophic forgetting indicated that sequential training led to significant forgetting, while single-session training with balanced batches and the MoE approach mitigated this issue. The MoE architecture preserved knowledge across multiple languages effectively. The research contributes open-sourced resources including the dataset (https://zenodo.org/doi/10.5281/zenodo.12677631), a balanced dataset creation tool (https://github.com/padas-lab-de/multi-language-dataset-creator), and the research codebase (https://github.com/ModMaamari/mixture-modular-experts).

Interpretability of deep reinforcement learning systems could assist operators with understanding how they interact with their environment. Vector quantization methods -- also called codebook methods -- discretize a neural network's latent space that is often suggested to yield emergent interpretability. We investigate whether vector quantization in fact provides interpretability in model-based reinforcement learning. Our experiments, conducted in the reinforcement learning environment Crafter, show that the codes of vector quantization models are inconsistent, have no guarantee of uniqueness, and have a limited impact on concept disentanglement, all of which are necessary traits for interpretability. We share insights on why vector quantization may be fundamentally insufficient for model interpretability.