One type of question that is commonly found in day-to-day scenarios is ``fan-out'' questions, complex multi-hop, multi-document reasoning questions that require finding information about a large number of entities. However, there exist few resources to evaluate this type of question-answering capability among large language models. To evaluate complex reasoning in LLMs more fully, we present FanOutQA, a high-quality dataset of fan-out question-answer pairs and human-annotated decompositions with English Wikipedia as the knowledge base. We formulate three benchmark settings across our dataset and benchmark 7 LLMs, including GPT-4, LLaMA 2, Claude-2.1, and Mixtral-8x7B, finding that contemporary models still have room to improve reasoning over inter-document dependencies in a long context. We provide our dataset and open-source tools to run models to encourage evaluation at https://fanoutqa.com

In-Context Learning (ICL) has been a powerful emergent property of large language models that has attracted increasing attention in recent years. In contrast to regular gradient-based learning, ICL is highly interpretable and does not require parameter updates. In this paper, we show that, for linearized transformer networks, ICL can be made explicit and permanent through the inclusion of bias terms. We mathematically demonstrate the equivalence between a model with ICL demonstration prompts and the same model with the additional bias terms. Our algorithm (ICLCA) allows for exact conversion in an inexpensive manner. Existing methods are not exact and require expensive parameter updates. We demonstrate the efficacy of our approach through experiments that show the exact incorporation of ICL tokens into a linear transformer. We further suggest how our method can be adapted to achieve cheap approximate conversion of ICL tokens, even in regular transformer networks that are not linearized. Our experiments on GPT-2 show that, even though the conversion is only approximate, the model still gains valuable context from the included bias terms.

In-context learning (ICL) empowers large language models (LLMs) to tackle new tasks by using a series of training instances as prompts. Since generating the prompts needs to sample from a vast pool of instances and annotate them (e.g., add labels in classification task), existing methods have proposed to select a subset of unlabeled examples for annotation, thus enhancing the quality of prompts and concurrently mitigating annotation costs. However, these methods often require a long time to select instances due to their complexity, hindering their practical viability. To address this limitation, we propose a graph-based selection method, FastGAS, designed to efficiently identify high-quality instances while minimizing computational overhead. Initially, we construct a data similarity graph based on instance similarities. Subsequently, employing a graph partitioning algorithm, we partition the graph into pieces. Within each piece (i.e., subgraph), we adopt a greedy approach to pick the most representative nodes. By aggregating nodes from diverse pieces and annotating the corresponding instances, we identify a set of diverse and representative instances for ICL. Compared to prior approaches, our method not only exhibits superior performance on different tasks but also significantly reduces selection time. In addition, we demonstrate the efficacy of our approach in LLMs of larger sizes.

One of the principal objectives of Natural Language Processing (NLP) is to generate meaningful representations from text. Improving the informativeness of the representations has led to a tremendous rise in the dimensionality and the memory footprint. It leads to a cascading effect amplifying the complexity of the downstream model by increasing its parameters. The available techniques cannot be applied to cross-modal applications such as text-to-image. To ameliorate these issues, a novel Text-to-Image methodology for generating fixed-length representations through a self-supervised Variational Auto-Encoder (VAE) for semantic evaluation applying transformers (TexIm FAST) has been proposed in this paper. The pictorial representations allow oblivious inference while retaining the linguistic intricacies, and are potent in cross-modal applications. TexIm FAST deals with variable-length sequences and generates fixed-length representations with over 75% reduced memory footprint. It enhances the efficiency of the models for downstream tasks by reducing its parameters. The efficacy of TexIm FAST has been extensively analyzed for the task of Semantic Textual Similarity (STS) upon the MSRPC, CNN/ Daily Mail, and XSum data-sets. The results demonstrate 6% improvement in accuracy compared to the baseline and showcase its exceptional ability to compare disparate length sequences such as a text with its summary.

Driven by the ever-growing volume and decentralized nature of data, coupled with the need to harness this data and generate knowledge from it, has led to the extensive use of distributed deep learning (DDL) techniques for training. These techniques rely on local training that is performed at the distributed nodes based on locally collected data, followed by a periodic synchronization process that combines these models to create a global model. However, frequent synchronization of DL models, encompassing millions to many billions of parameters, creates a communication bottleneck, severely hindering scalability. Worse yet, DDL algorithms typically waste valuable bandwidth, and make themselves less practical in bandwidth-constrained federated settings, by relying on overly simplistic, periodic, and rigid synchronization schedules. These drawbacks also have a direct impact on the time required for the training process, necessitating excessive time for data communication. To address these shortcomings, we propose Federated Dynamic Averaging (FDA), a communication-efficient DDL strategy that dynamically triggers synchronization based on the value of the model variance. In essence, the costly synchronization step is triggered only if the local models, which are initialized from a common global model after each synchronization, have significantly diverged. This decision is facilitated by the communication of a small local state from each distributed node/worker. Through extensive experiments across a wide range of learning tasks we demonstrate that FDA reduces communication cost by orders of magnitude, compared to both traditional and cutting-edge communication-efficient algorithms. Additionally, we show that FDA maintains robust performance across diverse data heterogeneity settings.

The increasing complexity of foundational models underscores the necessity for explainability, particularly for fine-tuning, the most widely used training method for adapting models to downstream tasks. Instance attribution, one type of explanation, attributes the model prediction to each training example by an instance score. However, the robustness of instance scores, specifically towards dataset resampling, has been overlooked. To bridge this gap, we propose a notion of robustness on the sign of the instance score. We theoretically and empirically demonstrate that the popular leave-one-out-based methods lack robustness, while the Shapley value behaves significantly better, but at a higher computational cost. Accordingly, we introduce an efficient fine-tuning-free approximation of the Shapley value (FreeShap) for instance attribution based on the neural tangent kernel. We empirically demonstrate that FreeShap outperforms other methods for instance attribution and other data-centric applications such as data removal, data selection, and wrong label detection, and further generalize our scale to large language models (LLMs). Our code is available at https://github.com/JTWang2000/FreeShap.

Advances in deep learning systems have allowed large models to match or surpass human accuracy on a number of skills such as image classification, basic programming, and standardized test taking. As the performance of the most capable models begin to saturate on tasks where humans already achieve high accuracy, it becomes necessary to benchmark models on increasingly complex abilities. One such task is forecasting the future outcome of events. In this work we describe experiments using a novel dataset of real world events and associated human predictions, an evaluation metric to measure forecasting ability, and the accuracy of a number of different LLM based forecasting designs on the provided dataset. Additionally, we analyze the performance of the LLM forecasters against human predictions and find that models still struggle to make accurate predictions about the future. Our follow-up experiments indicate this is likely due to models' tendency to guess that most events are unlikely to occur (which tends to be true for many prediction datasets, but does not reflect actual forecasting abilities). We reflect on next steps for developing a systematic and reliable approach to studying LLM forecasting.

Conformal Prediction (CP) is a popular uncertainty quantification method that provides distribution-free, statistically valid prediction sets, assuming that training and test data are exchangeable. In such a case, CP's prediction sets are guaranteed to cover the (unknown) true test output with a user-specified probability. Nevertheless, this guarantee is violated when the data is subjected to adversarial attacks, which often result in a significant loss of coverage. Recently, several approaches have been put forward to recover CP guarantees in this setting. These approaches leverage variations of randomised smoothing to produce conservative sets which account for the effect of the adversarial perturbations. They are, however, limited in that they only support $\ell^2$-bounded perturbations and classification tasks. This paper introduces VRCP (Verifiably Robust Conformal Prediction), a new framework that leverages recent neural network verification methods to recover coverage guarantees under adversarial attacks. Our VRCP method is the first to support perturbations bounded by arbitrary norms including $\ell^1$, $\ell^2$, and $\ell^\infty$, as well as regression tasks. We evaluate and compare our approach on image classification tasks (CIFAR10, CIFAR100, and TinyImageNet) and regression tasks for deep reinforcement learning environments. In every case, VRCP achieves above nominal coverage and yields significantly more efficient and informative prediction regions than the SotA.

Satellite-based remote sensing has revolutionised the way we address global challenges in a rapidly evolving world. Huge quantities of Earth Observation (EO) data are generated by satellite sensors daily, but processing these large datasets for use in ML pipelines is technically and computationally challenging. Specifically, different types of EO data are often hosted on a variety of platforms, with differing availability for Python preprocessing tools. In addition, spatial alignment across data sources and data tiling can present significant technical hurdles for novice users. While some preprocessed EO datasets exist, their content is often limited to optical or near-optical wavelength data, which is ineffective at night or in adverse weather conditions. Synthetic Aperture Radar (SAR), an active sensing technique based on microwave length radiation, offers a viable alternative. However, the application of machine learning to SAR has been limited due to a lack of ML-ready data and pipelines, particularly for the full diversity of SAR data, including polarimetry, coherence and interferometry. We introduce M3LEO, a multi-modal, multi-label EO dataset that includes polarimetric, interferometric, and coherence SAR data derived from Sentinel-1, alongside Sentinel-2 RGB imagery and a suite of labelled tasks for model evaluation. M3LEO spans 17.5TB and contains approximately 10M data chips across six geographic regions. The dataset is complemented by a flexible PyTorch Lightning framework, with configuration management using Hydra. We provide tools to process any dataset available on popular platforms such as Google Earth Engine for integration with our framework. Initial experiments validate the utility of our data and framework, showing that SAR imagery contains information additional to that extractable from RGB data. Data at huggingface.co/M3LEO, and code at github.com/spaceml-org/M3LEO.

Large language models (LLMs) can generate long-form and coherent text, yet they often hallucinate facts, which undermines their reliability. To mitigate this issue, inference-time methods steer LLM representations toward the "truthful directions" previously learned for truth elicitation. However, applying these truthful directions with the same intensity fails to generalize across different query contexts. We propose LITO, a Learnable Intervention method for Truthfulness Optimization that automatically identifies the optimal intervention intensity tailored to each specific context. LITO explores a sequence of model generations based on increasing levels of intervention intensities. It selects the most accurate response or refuses to answer when the predictions are highly uncertain. Experiments on multiple LLMs and question-answering datasets demonstrate that LITO improves truthfulness while preserving task accuracy. The adaptive nature of LITO counters the limitations of one-size-fits-all intervention methods, maximizing truthfulness by reflecting the model's internal knowledge only when it is confident. Our code is available at https://github.com/launchnlp/LITO.