This study explores the impact of class outliers on the effectiveness of example-based explainability methods for black-box machine learning models. We reformulate existing explainability evaluation metrics, such as correctness and relevance, specifically for example-based methods, and introduce a new metric, distinguishability. Using these metrics, we highlight the shortcomings of current example-based explainability methods, including those who attempt to suppress class outliers. We conduct experiments on two datasets, a text classification dataset and an image classification dataset, and evaluate the performance of four state-of-the-art explainability methods. Our findings underscore the need for robust techniques to tackle the challenges posed by class outliers.
We propose selective debiasing -- an inference-time safety mechanism that aims to increase the overall quality of models in terms of prediction performance and fairness in the situation when re-training a model is prohibitive. The method is inspired by selective prediction, where some predictions that are considered low quality are discarded at inference time. In our approach, we identify the potentially biased model predictions and, instead of discarding them, we debias them using LEACE -- a post-processing debiasing method. To select problematic predictions, we propose a bias quantification approach based on KL divergence, which achieves better results than standard UQ methods. Experiments with text classification datasets demonstrate that selective debiasing helps to close the performance gap between post-processing methods and at-training and pre-processing debiasing techniques.
This is the first work to investigate the effectiveness of BERT-based contextual embeddings in active learning (AL) tasks on cold-start scenarios, where traditional fine-tuning is infeasible due to the absence of labeled data. Our primary contribution is the proposal of a more robust fine-tuning pipeline - DoTCAL - that diminishes the reliance on labeled data in AL using two steps: (1) fully leveraging unlabeled data through domain adaptation of the embeddings via masked language modeling and (2) further adjusting model weights using labeled data selected by AL. Our evaluation contrasts BERT-based embeddings with other prevalent text representation paradigms, including Bag of Words (BoW), Latent Semantic Indexing (LSI), and FastText, at two critical stages of the AL process: instance selection and classification. Experiments conducted on eight ATC benchmarks with varying AL budgets (number of labeled instances) and number of instances (about 5,000 to 300,000) demonstrate DoTCAL's superior effectiveness, achieving up to a 33% improvement in Macro-F1 while reducing labeling efforts by half compared to the traditional one-step method. We also found that in several tasks, BoW and LSI (due to information aggregation) produce results superior (up to 59% ) to BERT, especially in low-budget scenarios and hard-to-classify tasks, which is quite surprising.
For many use-cases, it is often important to explain the prediction of a black-box model by identifying the most influential training data samples. Existing approaches lack customization for user intent and often provide a homogeneous set of explanation samples, failing to reveal the model's reasoning from different angles. In this paper, we propose AIDE, an approach for providing antithetical (i.e., contrastive), intent-based, diverse explanations for opaque and complex models. AIDE distinguishes three types of explainability intents: interpreting a correct, investigating a wrong, and clarifying an ambiguous prediction. For each intent, AIDE selects an appropriate set of influential training samples that support or oppose the prediction either directly or by contrast. To provide a succinct summary, AIDE uses diversity-aware sampling to avoid redundancy and increase coverage of the training data. We demonstrate the effectiveness of AIDE on image and text classification tasks, in three ways: quantitatively, assessing correctness and continuity; qualitatively, comparing anecdotal evidence from AIDE and other example-based approaches; and via a user study, evaluating multiple aspects of AIDE. The results show that AIDE addresses the limitations of existing methods and exhibits desirable traits for an explainability method.
With the rapid advancement of artificial intelligence technologies in human life, particularly within the automotive industry, the popularity of smart vehicles has increased. Designing an intelligent car cabin capable of seamless interactions with both the driver and the vehicle emerges as a crucial solution to address human-vehicle interaction challenges. This project aims to implement a digital voice assistant that recognizes vehicle commands. It utilizes three main techniques: speech-to-text conversion, text classification, and calibrating the classifier in order to detect out-of-distribution (OOD) sentences. Using Vosk, an LLM model, voices in the vehicle environment are converted into text format. Then, after pre-processing the text, an SVM classifies them. Using Platt scaling the SVM classifier outputs become calibrated, which makes them probabilistic. The experimental result shows that the proposed model achieves command recognition with 96.43% accuracy, 96.83% precision, 96.43% recall, 96.39% F1-score, and 0.1437 cross-entropy loss. Furthermore, the optimal threshold for OOD sentences is 0.4.
Annotation and classification of legal text are central components of empirical legal research. Traditionally, these tasks are often delegated to trained research assistants. Motivated by the advances in language modeling, empirical legal scholars are increasingly turning to prompting commercial models, hoping that it will alleviate the significant cost of human annotation. Despite growing use, our understanding of how to best utilize large language models for legal tasks remains limited. We conduct a comprehensive study of 260 legal text classification tasks, nearly all new to the machine learning community. Starting from GPT-4 as a baseline, we show that it has non-trivial but highly varied zero-shot accuracy, often exhibiting performance that may be insufficient for legal work. We then demonstrate that a lightly fine-tuned Llama 3 model vastly outperforms GPT-4 on almost all tasks, typically by double-digit percentage points. We find that larger models respond better to fine-tuning than smaller models. A few tens to hundreds of examples suffice to achieve high classification accuracy. Notably, we can fine-tune a single model on all 260 tasks simultaneously at a small loss in accuracy relative to having a separate model for each task. Our work points to a viable alternative to the predominant practice of prompting commercial models. For concrete legal tasks with some available labeled data, researchers are better off using a fine-tuned open-source model.
This paper introduces a multi-level, multi-label text classification dataset comprising over 3000 documents. The dataset features literary and critical texts from 19th-century Ottoman Turkish and Russian. It is the first study to apply large language models (LLMs) to this dataset, sourced from prominent literary periodicals of the era. The texts have been meticulously organized and labeled. This was done according to a taxonomic framework that takes into account both their structural and semantic attributes. Articles are categorized and tagged with bibliometric metadata by human experts. We present baseline classification results using a classical bag-of-words (BoW) naive Bayes model and three modern LLMs: multilingual BERT, Falcon, and Llama-v2. We found that in certain cases, Bag of Words (BoW) outperforms Large Language Models (LLMs), emphasizing the need for additional research, especially in low-resource language settings. This dataset is expected to be a valuable resource for researchers in natural language processing and machine learning, especially for historical and low-resource languages. The dataset is publicly available^1.
Few-shot action recognition aims to address the high cost and impracticality of manually labeling complex and variable video data in action recognition. It requires accurately classifying human actions in videos using only a few labeled examples per class. Compared to few-shot learning in image scenarios, few-shot action recognition is more challenging due to the intrinsic complexity of video data. Numerous approaches have driven significant advancements in few-shot action recognition, which underscores the need for a comprehensive survey. Unlike early surveys that focus on few-shot image or text classification, we deeply consider the unique challenges of few-shot action recognition. In this survey, we provide a comprehensive review of recent methods and introduce a novel and systematic taxonomy of existing approaches, accompanied by a detailed analysis. We categorize the methods into generative-based and meta-learning frameworks, and further elaborate on the methods within the meta-learning framework, covering aspects: video instance representation, category prototype learning, and generalized video alignment. Additionally, the survey presents the commonly used benchmarks and discusses relevant advanced topics and promising future directions. We hope this survey can serve as a valuable resource for researchers, offering essential guidance to newcomers and stimulating seasoned researchers with fresh insights.
With the advent of foundation models, prompt tuning has positioned itself as an important technique for directing model behaviors and eliciting desired responses. Prompt tuning regards selecting appropriate keywords included into the input, thereby adapting to the downstream task without adjusting or fine-tuning the model parameters. There is a wide range of work in prompt tuning, from approaches that directly harness the backpropagated gradient signals from the model, to those employing black-box optimization such as reinforcement learning (RL) methods. Our primary focus is on RLPrompt, which aims to find optimal prompt tokens leveraging soft Q-learning. While the results show promise, we have observed that the prompts frequently appear unnatural, which impedes their interpretability. We address this limitation by using sparse Tsallis entropy regularization, a principled approach to filtering out unlikely tokens from consideration. We extensively evaluate our approach across various tasks, including few-shot text classification, unsupervised text style transfer, and textual inversion from images. The results indicate a notable improvement over baselines, highlighting the efficacy of our approach in addressing the challenges of prompt tuning. Moreover, we show that the prompts discovered using our method are more natural and interpretable compared to those from other baselines.
Using Large Language Models (LLMs) to generate synthetic data for model training has become increasingly popular in recent years. While LLMs are capable of producing realistic training data, the effectiveness of data generation is influenced by various factors, including the choice of prompt, task complexity, and the quality, quantity, and diversity of the generated data. In this work, we focus exclusively on using synthetic data for text classification tasks. Specifically, we use natural language understanding (NLU) models trained on synthetic data to assess the quality of synthetic data from different generation approaches. This work provides an empirical analysis of the impact of these factors and offers recommendations for better data generation practices.
The success of pretrained language models (PLMs) across a spate of use-cases has led to significant investment from the NLP community towards building domain-specific foundational models. On the other hand, in mission critical settings such as biomedical applications, other aspects also factor in-chief of which is a model's ability to produce reasonable estimates of its own uncertainty. In the present study, we discuss these two desiderata through the lens of how they shape the entropy of a model's output probability distribution. We find that domain specificity and uncertainty awareness can often be successfully combined, but the exact task at hand weighs in much more strongly.
We study the problem of continual test-time adaption where the goal is to adapt a source pre-trained model to a sequence of unlabelled target domains at test time. Existing methods on test-time training suffer from several limitations: (1) Mismatch between the feature extractor and classifier; (2) Interference between the main and self-supervised tasks; (3) Lack of the ability to quickly adapt to the current distribution. In light of these challenges, we propose a cascading paradigm that simultaneously updates the feature extractor and classifier at test time, mitigating the mismatch between them and enabling long-term model adaptation. The pre-training of our model is structured within a meta-learning framework, thereby minimizing the interference between the main and self-supervised tasks and encouraging fast adaptation in the presence of limited unlabelled data. Additionally, we introduce innovative evaluation metrics, average accuracy and forward transfer, to effectively measure the model's adaptation capabilities in dynamic, real-world scenarios. Extensive experiments and ablation studies demonstrate the superiority of our approach in a range of tasks including image classification, text classification, and speech recognition.
We introduce NOVIC, an innovative real-time uNconstrained Open Vocabulary Image Classifier that uses an autoregressive transformer to generatively output classification labels as language. Leveraging the extensive knowledge of CLIP models, NOVIC harnesses the embedding space to enable zero-shot transfer from pure text to images. Traditional CLIP models, despite their ability for open vocabulary classification, require an exhaustive prompt of potential class labels, restricting their application to images of known content or context. To address this, we propose an "object decoder" model that is trained on a large-scale 92M-target dataset of templated object noun sets and LLM-generated captions to always output the object noun in question. This effectively inverts the CLIP text encoder and allows textual object labels from essentially the entire English language to be generated directly from image-derived embedding vectors, without requiring any a priori knowledge of the potential content of an image, and without any label biases. The trained decoders are tested on a mix of manually and web-curated datasets, as well as standard image classification benchmarks, and achieve fine-grained prompt-free prediction scores of up to 87.5%, a strong result considering the model must work for any conceivable image and without any contextual clues.
Average hold time is a concern for call centers because it affects customer satisfaction. Contact centers should instruct their agents to use special on-hold scripts to maintain positive interactions with clients. This study presents a natural language processing model that detects on-hold phrases in customer service calls transcribed by automatic speech recognition technology. The task of finding hold scripts in dialogue was formulated as a multiclass text classification problem with three mutually exclusive classes: scripts for putting a client on hold, scripts for returning to a client, and phrases irrelevant to on-hold scripts. We collected an in-house dataset of calls and labeled each dialogue turn in each call. We fine-tuned RuBERT on the dataset by exploring various hyperparameter sets and achieved high model performance. The developed model can help agent monitoring by providing a way to check whether an agent follows predefined on-hold scripts.
Recently, various pre-trained language models (PLMs) have been proposed to prove their impressive performances on a wide range of few-shot tasks. However, limited by the unstructured prior knowledge in PLMs, it is difficult to maintain consistent performance on complex structured scenarios, such as hierarchical text classification (HTC), especially when the downstream data is extremely scarce. The main challenge is how to transfer the unstructured semantic space in PLMs to the downstream domain hierarchy. Unlike previous work on HTC which directly performs multi-label classification or uses graph neural network (GNN) to inject label hierarchy, in this work, we study the HTC problem under a few-shot setting to adapt knowledge in PLMs from an unstructured manner to the downstream hierarchy. Technically, we design a simple yet effective method named Hierarchical Iterative Conditional Random Field (HierICRF) to search the most domain-challenging directions and exquisitely crafts domain-hierarchy adaptation as a hierarchical iterative language modeling problem, and then it encourages the model to make hierarchical consistency self-correction during the inference, thereby achieving knowledge transfer with hierarchical consistency preservation. We perform HierICRF on various architectures, and extensive experiments on two popular HTC datasets demonstrate that prompt with HierICRF significantly boosts the few-shot HTC performance with an average Micro-F1 by 28.80% to 1.50% and Macro-F1 by 36.29% to 1.5% over the previous state-of-the-art (SOTA) baselines under few-shot settings, while remaining SOTA hierarchical consistency performance.
Deep learning has been increasingly incorporated into various computational pathology applications to improve its efficiency, accuracy, and robustness. Although successful, most previous approaches for image classification have crucial drawbacks. There exist numerous tasks in pathology, but one needs to build a model per task, i.e., a task-specific model, thereby increasing the number of models, training resources, and cost. Moreover, transferring arbitrary task-specific model to another task is still a challenging problem. Herein, we propose a task-agnostic generative and general pathology image classifier, so called GPC, that aims at learning from diverse kinds of pathology images and conducting numerous classification tasks in a unified model. GPC, equipped with a convolutional neural network and a Transformer-based language model, maps pathology images into a high-dimensional feature space and generates pertinent class labels as texts via the image-to-text classification mechanism. We evaluate GPC on six datasets for four different pathology image classification tasks. Experimental results show that GPC holds considerable potential for developing an effective and efficient universal model for pathology image analysis.
Supervised machine-learning models for predicting user behavior offer a challenging classification problem with lower average prediction performance scores than other text classification tasks. This study evaluates multi-task learning frameworks grounded in Cognitive Appraisal Theory to predict user behavior as a function of users' self-expression and psychological attributes. Our experiments show that users' language and traits improve predictions above and beyond models predicting only from text. Our findings highlight the importance of integrating psychological constructs into NLP to enhance the understanding and prediction of user actions. We close with a discussion of the implications for future applications of large language models for computational psychology.
Online platform businesses can be identified by using web-scraped texts. This is a classification problem that combines elements of natural language processing and rare event detection. Because online platforms are rare, accurately identifying them with Machine Learning algorithms is challenging. Here, we describe the development of a Machine Learning-based text classification approach that reduces the number of false positives as much as possible. It greatly reduces the bias in the estimates obtained by using calibrated probabilities and ensembles.
Existing research on learning with noisy labels predominantly focuses on synthetic label noise. Although synthetic noise possesses well-defined structural properties, it often fails to accurately replicate real-world noise patterns. In recent years, there has been a concerted effort to construct generalizable and controllable instance-dependent noise datasets for image classification, significantly advancing the development of noise-robust learning in this area. However, studies on noisy label learning for text classification remain scarce. To better understand label noise in real-world text classification settings, we constructed the benchmark dataset NoisyAG-News through manual annotation. Initially, we analyzed the annotated data to gather observations about real-world noise. We qualitatively and quantitatively demonstrated that real-world noisy labels adhere to instance-dependent patterns. Subsequently, we conducted comprehensive learning experiments on NoisyAG-News and its corresponding synthetic noise datasets using pre-trained language models and noise-handling techniques. Our findings reveal that while pre-trained models are resilient to synthetic noise, they struggle against instance-dependent noise, with samples of varying confusion levels showing inconsistent performance during training and testing. These real-world noise patterns pose new, significant challenges, prompting a reevaluation of noisy label handling methods. We hope that NoisyAG-News will facilitate the development and evaluation of future solutions for learning with noisy labels.
We study open-world multi-label text classification under extremely weak supervision (XWS), where the user only provides a brief description for classification objectives without any labels or ground-truth label space. Similar single-label XWS settings have been explored recently, however, these methods cannot be easily adapted for multi-label. We observe that (1) most documents have a dominant class covering the majority of content and (2) long-tail labels would appear in some documents as a dominant class. Therefore, we first utilize the user description to prompt a large language model (LLM) for dominant keyphrases of a subset of raw documents, and then construct a (initial) label space via clustering. We further apply a zero-shot multi-label classifier to locate the documents with small top predicted scores, so we can revisit their dominant keyphrases for more long-tail labels. We iterate this process to discover a comprehensive label space and construct a multi-label classifier as a novel method, X-MLClass. X-MLClass exhibits a remarkable increase in ground-truth label space coverage on various datasets, for example, a 40% improvement on the AAPD dataset over topic modeling and keyword extraction methods. Moreover, X-MLClass achieves the best end-to-end multi-label classification accuracy.