In-context learning (ICL) adapts Large Language Models (LLMs) to new tasks, without requiring any parameter updates, but few annotated examples as input. In this work, we investigate selective annotation for ICL, where there is a limited budget for annotating examples, similar to low-budget active learning (AL). Although uncertainty-based selection is unreliable with few annotated data, we present CoverICL, an adaptive graph-based selection algorithm, that effectively incorporates uncertainty sampling into selective annotation for ICL. First, CoverICL builds a nearest-neighbor graph based on the semantic similarity between candidate ICL examples. Then, CoverICL employs uncertainty estimation by the LLM to identify hard examples for the task. Selective annotation is performed over the active graph of the hard examples, adapting the process to the particular LLM used and the task tackled. CoverICL selects the most representative examples by solving a Maximum Coverage problem, approximating diversity-based sampling. Extensive experiments on ten datasets and seven LLMs show that, by incorporating uncertainty via coverage on the active graph, CoverICL (1) outperforms existing AL methods for ICL by 2–4.6% accuracy points, (2) is up to 2x more budget-efficient than SOTA methods for low-budget AL, and (3) generalizes better across tasks compared to non-graph alternatives.

Effectively training language models on longinputs poses many technical challenges. As acost consideration, languages models are pre-trained on a fixed sequence length before beingadapted to longer sequences. We explore var-ious methods for adapting models to longerinputs by training on segmented sequences andan interpolation-based method for extendingabsolute positional embeddings. We developa training procedure to extend the input con-text size of pretrained models with no architec-tural changes and no additional memory coststhan training on the original input lengths. Bysub-sampling segments from long inputs whilemaintaining their original position the model isable to learn new positional interactions. Ourmethod benefits both models trained with abso-lute positional embeddings, by extending theirinput contexts, as well as popular relative posi-tional embedding methods showing a reducedperplexity on sequences longer than they weretrained on. We demonstrate our method canextend input contexts by a factor of 4× whileimproving perplexity.

Citation intention Classification (CIC) tools classify citations by their intention (e.g., background, motivation) and assist readers in evaluating the contribution of scientific literature. Prior research has shown that pretrained language models (PLMs) such as SciBERT can achieve state-of-the-art performance on CIC benchmarks. PLMs are trained via self-supervision tasks on a large corpus of general text and can quickly adapt to CIC tasks via moderate fine-tuning on the corresponding dataset. Despite their advantages, PLMs can easily overfit small datasets during fine-tuning. In this paper, we propose a multi-task learning (MTL) framework that jointly fine-tunes PLMs on a dataset of primary interest together with multiple auxiliary CIC datasets to take advantage of additional supervision signals. We develop a data-driven task relation learning (TRL) method that controls the contribution of auxiliary datasets to avoid negative transfer and expensive hyper-parameter tuning. We conduct experiments on three CIC datasets and show that fine-tuning with additional datasets can improve the PLMs’ generalization performance on the primary dataset. PLMs fine-tuned with our proposed framework outperform the current state-of-the-art models by 7% to 11% on small datasets while aligning with the best-performing model on a large dataset.

Deep learning-based Natural Language Processing (NLP) models are vulnerable to adversarial attacks, where small perturbations can cause a model to misclassify. Adversarial Training (AT) is often used to increase model robustness. However, we have discovered an intriguing phenomenon: deliberately or accidentally miscalibrating models masks gradients in a way that interferes with adversarial attack search methods, giving rise to an apparent increase in robustness. We show that this observed gain in robustness is an illusion of robustness (IOR), and demonstrate how an adversary can perform various forms of test-time temperature calibration to nullify the aforementioned interference and allow the adversarial attack to find adversarial examples. Hence, we urge the NLP community to incorporate test-time temperature scaling into their robustness evaluations to ensure that any observed gains are genuine. Finally, we show how the temperature can be scaled during training to improve genuine robustness.

Large Language Models (LLMs) frequently hallucinate, impeding their reliability in mission-critical situations. One approach to address this issue is to provide citations to relevant sources alongside generated content, enhancing the verifiability of generations. However, citing passages accurately in answers remains a substantial challenge. This paper proposes a weakly-supervised fine-tuning method leveraging factual consistency models (FCMs). Our approach alternates between generating texts with citations and supervised fine-tuning with FCM-filtered citation data. Focused learning is integrated into the objective, directing the fine-tuning process to emphasise the factual unit tokens, as measured by an FCM. Results on the ALCE few-shot citation benchmark with various instruction-tuned LLMs demonstrate superior performance compared to in-context learning, vanilla supervised fine-tuning, and state-of-the-art methods, with an average improvement of 34.1, 15.5, and 10.5 citation F₁ points, respectively. Moreover, in a domain transfer setting we show that the obtained citation generation ability robustly transfers to unseen datasets. Notably, our citation improvements contribute to the lowest factual error rate across baselines.

Given a data lake of tabular data as well as a query table, how can we retrieve all the tables in the data lake that can be unioned with the query table? Table union search constitutes an essential task in data discovery and preparation as it enables data scientists to navigate massive open data repositories. Existing methods identify uniability based on column representations (word surface forms or token embeddings) and column relation represented by column representation similarity. However, the semantic similarity obtained between column representations is often insufficient to reveal latent relational features to describe the column relation between pair of columns and not robust to the table noise. To address these issues, in this paper, we propose a multi-stage self-supervised table union search framework called AutoTUS, which represents column relation as a vector– column relational representation and learn column relational representation in a multi-stage manner that can better describe column relation for unionability prediction. In particular, the large language model powered contextualized column relation encoder is updated by adaptive clustering and pseudo label classification iteratively so that the better column relational representation can be learned. Moreover, to improve the robustness of the model against table noises, we propose table noise generator to add table noise to the training table data. Experiments on real-world datasets as well as synthetic test set augmented with table noise show that AutoTUS achieves 5.2% performance gain over the SOTA baseline.

Recent advances in large language models have revolutionized many sectors, including the database industry. One common challenge when dealing with large volumes of tabular data is the pervasive use of abbreviated column names, which can negatively impact performance on various data search, access, and understanding tasks. To address this issue, we introduce a new task, called NameGuess, to expand column names (used in database schema) as a natural language generation problem. We create a training dataset of 384K abbreviated-expanded column pairs using a new data fabrication method and a human-annotated evaluation benchmark that includes 9.2K examples from real-world tables. To tackle the complexities associated with polysemy and ambiguity in NameGuess, we enhance auto-regressive language models by conditioning on table content and column header names – yielding a fine-tuned model (with 2.7B parameters) that matches human performance. Furthermore, we conduct a comprehensive analysis (on multiple LLMs) to validate the effectiveness of table content in NameGuess and identify promising future opportunities. Code has been made available at https://github.com/amazon-science/nameguess.

Recent work has found that multi-task training with a large number of diverse tasks can uniformly improve downstream performance on unseen target tasks. In contrast, literature on task transferability has established that the choice of intermediate tasks can heavily affect downstream task performance. In this work, we aim to disentangle the effect of scale and relatedness of tasks in multi-task representation learning. We find that, on average, increasing the scale of multi-task learning, in terms of the number of tasks, indeed results in better learned representations than smaller multi-task setups. However, if the target tasks are known ahead of time, then training on a smaller set of related tasks is competitive to the large-scale multi-task training at a reduced computational cost.

The goal of meta-learning is to learn to adapt to a new task with only a few labeled examples. Inspired by the recent progress in large language models, we propose in-context tuning (ICT), which recasts task adaptation and prediction as a simple sequence prediction problem: to form the input sequence, we concatenate the task instruction, labeled in-context examples, and the target input to predict; to meta-train the model to learn from in-context examples, we fine-tune a pre-trained language model (LM) to predict the target label given the input sequence on a collection of tasks.We benchmark our method on two collections of text classification tasks: LAMA and BinaryClfs. Compared to MAML which adapts the model through gradient descent, our method leverages the inductive bias of pre-trained LMs to perform pattern matching, and outperforms MAML by an absolute 6% average AUC-ROC score on BinaryClfs, gaining more advantage with increasing model size. Compared to non-fine-tuned in-context learning (i.e. prompting a raw LM), in-context tuning meta-trains the model to learn from in-context examples. On BinaryClfs, ICT improves the average AUC-ROC score by an absolute 10%, and reduces the variance due to example ordering by 6x and example choices by 2x.

Knowledge Graph Question Answering (KGQA) involves retrieving entities as answers from a Knowledge Graph (KG) using natural language queries. The challenge is to learn to reason over question-relevant KG facts that traverse KG entities and lead to the question answers. To facilitate reasoning, the question is decoded into instructions, which are dense question representations used to guide the KG traversals. However, if the derived instructions do not exactly match the underlying KG information, they may lead to reasoning under irrelevant context.Our method, termed ReaRev, introduces a new way to KGQA reasoning with respectto both instruction decoding and execution. To improve instruction decoding, we perform reasoning in an adaptive manner, where KG-aware information is used to iteratively update the initial instructions. To improve instruction execution, we emulate breadth-first search (BFS) with graph neural networks (GNNs). The BFS strategy treats the instructions as a set and allows our method to decide on their execution order on the fly. Experimental results on three KGQA benchmarks demonstrate the ReaRev’s effectiveness compared with previous state-of-the-art, especially when the KG is incomplete or when we tackle complex questions. Our code is publicly available at https://github.com/cmavro/ReaRev_KGQA.

Knowledge Distillation (KD) offers a natural way to reduce the latency and memory/energy usage of massive pretrained models that have come to dominate Natural Language Processing (NLP) in recent years. While numerous sophisticated variants of KD algorithms have been proposed for NLP applications, the key factors underpinning the optimal distillation performance are often confounded and remain unclear. We aim to identify how different components in the KD pipeline affect the resulting performance and how much the optimal KD pipeline varies across different datasets/tasks, such as the data augmentation policy, the loss function, and the intermediate representation for transferring the knowledge between teacher and student. To tease apart their effects, we propose Distiller, a meta KD framework that systematically combines a broad range of techniques across different stages of the KD pipeline, which enables us to quantify each component’s contribution. Within Distiller, we unify commonly used objectives for distillation of intermediate representations under a universal mutual information (MI) objective and propose a class of MI-objective functions with better bias/variance trade-off for estimating the MI between the teacher and the student. On a diverse set of NLP datasets, the best Distiller configurations are identified via large-scale hyper-parameter optimization. Our experiments reveal the following: 1) the approach used to distill the intermediate representations is the most important factor in KD performance, 2) among different objectives for intermediate distillation, MI-performs the best, and 3) data augmentation provides a large boost for small training datasets or small student networks. Moreover, we find that different datasets/tasks prefer different KD algorithms, and thus propose a simple AutoDistiller algorithm that can recommend a good KD pipeline for a new dataset.

Quantitatively measuring the impact-related aspects of scientific, engineering, and technological (SET) innovations is a fundamental problem with broad applications. Traditional citation-based measures for assessing the impact of innovations and related entities do not take into account the content of the publications. This limits their ability to provide rigorous quality-related metrics because they cannot account for the reasons that led to a citation. We present approaches to estimate content-aware bibliometrics to quantitatively measure the scholarly impact of a publication. Our approaches assess the impact of a cited publication by the extent to which the cited publication informs the citing publication. We introduce a new metric, called “Content Informed Index” (CII), that uses the content of the paper as a source of distant-supervision, to quantify how much the cited-node informs the citing-node. We evaluate the weights estimated by our approach on three manually annotated datasets, where the annotations quantify the extent of information in the citation. Particularly, we evaluate how well the ranking imposed by our approach associates with the ranking imposed by the manual annotations. CII achieves up to 103% improvement in performance as compared to the second-best performing approach.