NLP research has explored different neural model architectures and sizes, datasets, training objectives, and transfer learning techniques. However, the choice of optimizer during training has not been explored as extensively. Typically, some variant of Stochastic Gradient Descent (SGD) is employed, selected among numerous variants, using unclear criteria, often with minimal or no tuning of the optimizer's hyperparameters. Experimenting with five GLUE datasets, two models (DistilBERT and DistilRoBERTa), and seven popular optimizers (SGD, SGD with Momentum, Adam, AdaMax, Nadam, AdamW, and AdaBound), we find that when the hyperparameters of the optimizers are tuned, there is no substantial difference in test performance across the five more elaborate (adaptive) optimizers, despite differences in training loss. Furthermore, tuning just the learning rate is in most cases as good as tuning all the hyperparameters. Hence, we recommend picking any of the best-behaved adaptive optimizers (e.g., Adam) and tuning only its learning rate. When no hyperparameter can be tuned, SGD with Momentum is the best choice.

We present an algorithmic solution to the problem of incremental belief updating in the context of Monte Carlo inference in Bayesian statistical models represented by probabilistic programs. Given a model and a sample-approximated posterior, our solution constructs a set of weighted observations to condition the model such that inference would result in the same posterior. This problem arises e.g. in multi-level modelling, incremental inference, inference in presence of privacy constraints. First, a set of virtual observations is selected, then, observation weights are found through a computationally efficient optimization procedure such that the reconstructed posterior coincides with or closely approximates the original posterior. We implement and apply the solution to a number of didactic examples and case studies, showing efficiency and robustness of our approach. The provided reference implementation is agnostic to the probabilistic programming language or the inference algorithm, and can be applied to most mainstream probabilistic programming environments.

In the rapidly evolving field of text generation, the demand for more precise control mechanisms has become increasingly apparent. To address this need, we present a novel methodology, LIFI, which offers a lightweight approach with fine-grained control for controlled text generation. Unlike previous studies that train pre-trained language models to follow discrete, categorical, and exclusive control codes, LIFI learns controlled text generation under the guidance of continuous, relative, and nonexclusive control codes. These fine-grained codes are automatically derived from an attribute classifier, initially trained with a small amount of labeled data and subsequently employed to label abundant unlabeled data, thus garnering more extensive supervision signals. Moreover, to achieve efficient control, we incorporate the fine-grained control codes with adapters, a parameter- and compute-efficient way to steer a pre-trained language model. We evaluate LIFI on two conventional tasks -- sentiment control and topic control -- and one newly proposed task -- stylistic novel writing. Comprehensive experimental results validate the effectiveness of our proposed methods, demonstrating substantial performance improvements over existing baselines.

The irreducible complexity of natural phenomena has led Graph Neural Networks to be employed as a standard model to perform representation learning tasks on graph-structured data. While their capacity to capture local and global patterns is remarkable, the implications associated with long-range and higher-order dependencies pose considerable challenges to such models. This work starts with a theoretical framework to reveal the impact of network's width, depth, and graph topology on the over-squashing phenomena in message-passing neural networks. Then, the work drifts towards, higher-order interactions and multi-relational inductive biases via Topological Neural Networks. Such models propagate messages through higher-dimensional structures, providing shortcuts or additional routes for information flow. With this construction, the underlying computational graph is no longer coupled with the input graph structure, thus mitigating the aforementioned bottlenecks while accounting also for higher-order interactions. Inspired by Graph Attention Networks, two topological attention networks are proposed: Simplicial and Cell Attention Networks. The rationale behind these architecture is to leverage the extended notion of neighbourhoods provided by the arrangement of groups of nodes within a simplicial or cell complex to design anisotropic aggregations able to measure the importance of the information coming from different regions of the domain. By doing so, they capture dependencies that conventional Graph Neural Networks might miss. Finally, a multi-way communication scheme is introduced with Enhanced Cellular Isomorphism Networks, which augment topological message passing schemes to enable a direct interactions among groups of nodes arranged in ring-like structures.

Massive open online courses (MOOCs) unlock the doors to free education for anyone around the globe with access to a computer and the internet. Despite this democratization of learning, the massive enrollment in these courses means it is almost impossible for one instructor to assess every student's writing assignment. As a result, peer grading, often guided by a straightforward rubric, is the method of choice. While convenient, peer grading often falls short in terms of reliability and validity. In this study, using 18 distinct settings, we explore the feasibility of leveraging large language models (LLMs) to replace peer grading in MOOCs. Specifically, we focus on two state-of-the-art LLMs: GPT-4 and GPT-3.5, across three distinct courses: Introductory Astronomy, Astrobiology, and the History and Philosophy of Astronomy. To instruct LLMs, we use three different prompts based on a variant of the zero-shot chain-of-thought (Zero-shot-CoT) prompting technique: Zero-shot-CoT combined with instructor-provided correct answers; Zero-shot-CoT in conjunction with both instructor-formulated answers and rubrics; and Zero-shot-CoT with instructor-offered correct answers and LLM-generated rubrics. Our results show that Zero-shot-CoT, when integrated with instructor-provided answers and rubrics, produces grades that are more aligned with those assigned by instructors compared to peer grading. However, the History and Philosophy of Astronomy course proves to be more challenging in terms of grading as opposed to other courses. Finally, our study reveals a promising direction for automating grading systems for MOOCs, especially in subjects with well-defined rubrics.

With the increasing need for inclusive and user-friendly technology, web accessibility is crucial to ensuring equal access to online content for individuals with disabilities, including visual, auditory, cognitive, or motor impairments. Despite the existence of accessibility guidelines and standards such as Web Content Accessibility Guidelines (WCAG) and the Web Accessibility Initiative (W3C), over 90% of websites still fail to meet the necessary accessibility requirements. For web users with disabilities, there exists a need for a tool to automatically fix web page accessibility errors. While research has demonstrated methods to find and target accessibility errors, no research has focused on effectively correcting such violations. This paper presents a novel approach to correcting accessibility violations on the web by modifying the document object model (DOM) in real time with foundation models. Leveraging accessibility error information, large language models (LLMs), and prompt engineering techniques, we achieved greater than a 51% reduction in accessibility violation errors after corrections on our novel benchmark: ACCESS. Our work demonstrates a valuable approach toward the direction of inclusive web content, and provides directions for future research to explore advanced methods to automate web accessibility.

Despite the advancements in large language models (LLMs) for mathematical reasoning, solving competition-level math problems remains a significant challenge, especially for open-source LLMs without external tools. We introduce the MMIQC dataset, comprising a mixture of processed web data and synthetic question-response pairs, aimed at enhancing the mathematical reasoning capabilities of base language models. Models fine-tuned on MMIQC consistently surpass their counterparts in performance on the MATH benchmark across various model sizes. Notably, Qwen-72B-MMIQC achieves a 45.0% accuracy, exceeding the previous open-source state-of-the-art by 8.2% and outperforming the initial version GPT-4 released in 2023. Extensive evaluation results on Hungarian high school finals suggest that such improvement can generalize to unseen data. Our ablation study on MMIQC reveals that a large part of the improvement can be attributed to our novel augmentation method, Iterative Question Composing (IQC), which involves iteratively composing new questions from seed problems using an LLM and applying rejection sampling through another LLM. The MMIQC dataset is available on the HuggingFace hub at https://huggingface.co/datasets/Vivacem/MMIQC. Our code is available at https://github.com/iiis-ai/IterativeQuestionComposing.

In multi-agent settings with mixed incentives, methods developed for zero-sum games have been shown to lead to detrimental outcomes. To address this issue, opponent shaping (OS) methods explicitly learn to influence the learning dynamics of co-players and empirically lead to improved individual and collective outcomes. However, OS methods have only been evaluated in low-dimensional environments due to the challenges associated with estimating higher-order derivatives or scaling model-free meta-learning. Alternative methods that scale to more complex settings either converge to undesirable solutions or rely on unrealistic assumptions about the environment or co-players. In this paper, we successfully scale an OS-based approach to general-sum games with temporally-extended actions and long-time horizons for the first time. After analysing the representations of the meta-state and history used by previous algorithms, we propose a simplified version called Shaper. We show empirically that Shaper leads to improved individual and collective outcomes in a range of challenging settings from literature. We further formalize a technique previously implicit in the literature, and analyse its contribution to opponent shaping. We show empirically that this technique is helpful for the functioning of prior methods in certain environments. Lastly, we show that previous environments, such as the CoinGame, are inadequate for analysing temporally-extended general-sum interactions.

We propose the Data Contamination Quiz (DCQ), a simple and effective approach to detect data contamination in large language models (LLMs) and estimate the amount of it. Specifically, we frame data contamination detection as a series of multiple-choice questions and devise a quiz format wherein three perturbed versions of each dataset instance are created. These changes only include word-level perturbations. The generated perturbed versions, along with the original instance, form the options in the DCQ, with an extra option accommodating the possibility that none of the provided choices is correct. Given that the only distinguishing signal among the choices is the exact wording relative to the original instance, an LLM, when tasked with identifying the original instance from the choices, gravitates towards the original one if it has been exposed to it in its pre-training phase--a trait intrinsic to LLMs. Tested over several datasets with GPT-4/3.5, our findings--while fully lacking access to LLMs' pre-training data and internal parameters--suggest that DCQ uncovers greater contamination levels compared to existing detection methods and proficiently bypasses more safety filters, especially those set to avoid generating copyrighted contents.

Researchers have slammed US officials for not rolling out stricter AI rules before popstar Taylor Swift became victim of deepfakes. Images showing the four-time Grammy winner in a series of sexual acts while dressed in Kansas City Chief memorabilia and in the stadium - and the pornography share - was viewed 47 million times online before being removed. A professor at George Washington University Law School said if proper legislation was'passed years ago' Swift and others would not have experienced such abuse. 'We are too little, too late at this point,' said Mary Anne Franks. 'It's not just going to be the 14-year-old girl or Taylor Swift. It's going to be politicians.