Auditing unwanted social bias in language models (LMs) is inherently hard due to the multidisciplinary nature of the work. In addition, the rapid evolution of LMs can make benchmarks irrelevant in no time. Bias auditing is further complicated by LM brittleness: when a presumably biased outcome is observed, is it due to model bias or model brittleness? We propose enlisting the models themselves to help construct bias auditing datasets that remain challenging, and introduce bias measures that distinguish between types of model errors. First, we extend an existing bias benchmark for NLI (BBNLI) using a combination of LM-generated lexical variations, adversarial filtering, and human validation. We demonstrate that the newly created dataset (BBNLInext) is more challenging than BBNLI: on average, BBNLI-next reduces the accuracy of state-of-the-art NLI models from 95.3%, as observed by BBNLI, to 58.6%. Second, we employ BBNLI-next to showcase the interplay between robustness and bias, and the subtlety in differentiating between the two. Third, we point out shortcomings in current bias scores used in the literature and propose bias measures that take into account pro-/anti-stereotype bias and model brittleness. We will publicly release the BBNLI-next dataset to inspire research on rapidly expanding benchmarks to keep up with model evolution, along with research on the robustness-bias interplay in bias auditing. Note: This paper contains offensive text examples.

Novel high-resolution pressure-sensor arrays allow treating pressure readings as standard images. Computer vision algorithms and methods such as Convolutional Neural Networks (CNN) can be used to identify contact objects. In this paper, a high-resolution tactile sensor has been attached to a robotic end-effector to identify contacted objects. Two CNN-based approaches have been employed to classify pressure images. These methods include a transfer learning approach using a pre-trained CNN on an RGB-images dataset and a custom-made CNN (TactNet) trained from scratch with tactile information. The transfer learning approach can be carried out by retraining the classification layers of the network or replacing these layers with an SVM. Overall, 11 configurations based on these methods have been tested: 8 transfer learning-based, and 3 TactNet-based. Moreover, a study of the performance of the methods and a comparative discussion with the current state-of-the-art on tactile object recognition is presented.

Large language models (LLMs) have shown great abilities of solving various natural language tasks in different domains. Due to the training objective of LLMs and their pretraining data, LLMs are not very well equipped for tasks involving structured data generation. We propose a framework, Prompting with Iterative Verification (PiVe), to improve graphbased generative capability of LLMs. We show how a small language model could be trained to act as a verifier module for the output of an LLM (i.e., ChatGPT), and to iteratively improve its performance via fine-grained corrective instructions. Additionally, we show how the verifier module could apply iterative corrections offline for a more cost-effective solution to the text-to-graph generation task. Experiments on three graph-based datasets show consistent improvement gained via PiVe. Additionally, we highlight how the proposed verifier module can be used as a data augmentation tool to help improve the quality of automatically generated parallel text-graph datasets. Our code and data are available at https://github.com/Jiuzhouh/PiVe.

Powered by advanced Artificial Intelligence (AI) techniques, conversational AI systems, such as ChatGPT and digital assistants like Siri, have been widely deployed in daily life. However, such systems may still produce content containing biases and stereotypes, causing potential social problems. Due to the data-driven, black-box nature of modern AI techniques, comprehensively identifying and measuring biases in conversational systems remains a challenging task. Particularly, it is hard to generate inputs that can comprehensively trigger potential bias due to the lack of data containing both social groups as well as biased properties. In addition, modern conversational systems can produce diverse responses (e.g., chatting and explanation), which makes existing bias detection methods simply based on the sentiment and the toxicity hardly being adopted. In this paper, we propose BiasAsker, an automated framework to identify and measure social bias in conversational AI systems. To obtain social groups and biased properties, we construct a comprehensive social bias dataset, containing a total of 841 groups and 8,110 biased properties. Given the dataset, BiasAsker automatically generates questions and adopts a novel method based on existence measurement to identify two types of biases (i.e., absolute bias and related bias) in conversational systems. Extensive experiments on 8 commercial systems and 2 famous research models, such as ChatGPT and GPT-3, show that 32.83% of the questions generated by BiasAsker can trigger biased behaviors in these widely deployed conversational systems. All the code, data, and experimental results have been released to facilitate future research.

This article investigates the effect of explicitly adding auxiliary algebraic trajectory information to neural networks for dynamical systems. We draw inspiration from the field of differential-algebraic equations and differential equations on manifolds and implement related methods in residual neural networks, despite some fundamental scenario differences. Constraint or auxiliary information effects are incorporated through stabilization as well as projection methods, and we show when to use which method based on experiments involving simulations of multi-body pendulums and molecular dynamics scenarios. Several of our methods are easy to implement in existing code and have limited impact on training performance while giving significant boosts in terms of inference.

Dialogue models are able to generate coherent and fluent responses, but they can still be challenging to control and may produce non-engaging, unsafe results. This unpredictability diminishes user trust and can hinder the use of the models in the real world. To address this, we introduce DialGuide, a novel framework for controlling dialogue model behavior using natural language rules, or guidelines. These guidelines provide information about the context they are applicable to and what should be included in the response, allowing the models to generate responses that are more closely aligned with the developer's expectations and intent. We evaluate DialGuide on three tasks in open-domain dialogue response generation: guideline selection, response generation, and response entailment verification. Our dataset contains 10,737 positive and 15,467 negative dialogue context-response-guideline triplets across two domains - chit-chat and safety. We provide baseline models for the tasks and benchmark their performance. We also demonstrate that DialGuide is effective in the dialogue safety domain, producing safe and engaging responses that follow developer guidelines.

Recent progress in large language models has enabled the deployment of many generative NLP applications. At the same time, it has also led to a misleading public discourse that ``it's all been solved.'' Not surprisingly, this has in turn made many NLP researchers -- especially those at the beginning of their career -- wonder about what NLP research area they should focus on. This document is a compilation of NLP research directions that are rich for exploration, reflecting the views of a diverse group of PhD students in an academic research lab. While we identify many research areas, many others exist; we do not cover those areas that are currently addressed by LLMs but where LLMs lag behind in performance, or those focused on LLM development. We welcome suggestions for other research directions to include: https://bit.ly/nlp-era-llm

Purpose: Self-supervised learning has been gaining attention in the medical field for its potential to improve computer-aided diagnosis. One popular method of self-supervised learning is masked image modeling (MIM), which involves masking a subset of input pixels and predicting the masked pixels. However, traditional MIM methods typically use a random masking strategy, which may not be ideal for medical images that often have a small region of interest for disease detection. To address this issue, this work aims to improve MIM for medical images and evaluate its effectiveness in an open X-ray image dataset. Methods: In this paper, we present a novel method called region-guided masked image modeling (RGMIM) for learning meaningful representation from X-ray images. Our method adopts a new masking strategy that utilizes organ mask information to identify valid regions for learning more meaningful representations. The proposed method was contrasted with five self-supervised learning techniques (MAE, SKD, Cross, BYOL, and, SimSiam). We conduct quantitative evaluations on an open lung X-ray image dataset as well as masking ratio hyperparameter studies. Results: When using the entire training set, RGMIM outperformed other comparable methods, achieving a 0.962 lung disease detection accuracy. Specifically, RGMIM significantly improved performance in small data volumes, such as 5% and 10% of the training set (846 and 1,693 images) compared to other methods, and achieved a 0.957 detection accuracy even when only 50% of the training set was used. Conclusions: RGMIM can mask more valid regions, facilitating the learning of discriminative representations and the subsequent high-accuracy lung disease detection. RGMIM outperforms other state-of-the-art self-supervised learning methods in experiments, particularly when limited training data is used.

Kai Zhang, PhD, Xiaoqian Jiang, PhD The University of Texas Health Science Center, McWilliams School of Biomedical Informatics, Houston, TX, USA Abstract: In the era of big data, there is an increasing need for healthcare providers, communities, and researchers to share data and collaborate to improve health outcomes, generate valuable insights, and advance research. The Health Insurance Portability and Accountability Act of 1996 (HIPAA) is a federal law designed to protect sensitive health information by defining regulations for protected health information (PHI). However, it does not provide efficient tools for detecting or removing PHI before data sharing. One of the challenges in this area of research is the heterogeneous nature of PHI fields in data across different parties. This variability makes rule-based sensitive variable identification systems that work on one database fail on another. To address this issue, our paper explores the use of machine learning algorithms to identify sensitive variables in structured data, thus facilitating the de-identification process. We made a key observation that the distributions of metadata of PHI fields and non-PHI fields are very different. Based on this novel finding, we engineered over 30 features from the metadata of the original features and used machine learning to build classification models to automatically identify PHI fields in structured Electronic Health Record (EHR) data. We trained the model on a variety of large EHR databases from different data sources and found that our algorithm achieves 99% accuracy when detecting PHI-related fields for unseen datasets.

Dialogue response generation requires an agent to generate a response according to the current dialogue history, in terms of which two-party dialogues have been well studied, but leaving a great gap for multi-party dialogues at the same time. Different from two-party dialogues where each response is a direct reply to its previous utterance, the addressee of a response utterance should be specified before it is generated in the multi-party scenario. Thanks to the huge amount of two-party conversational data, various pre-trained language models for two-party dialogue response generation have been proposed. However, due to the lack of annotated addressee labels in multi-party dialogue datasets, it is hard to use them to pre-train a response generation model for multi-party dialogues. To tackle this obstacle, we propose an Expectation-Maximization (EM) approach that iteratively performs the expectation steps to generate addressee labels, and the maximization steps to optimize a response generation model. Theoretical analyses and extensive experiments have justified the feasibility and effectiveness of our proposed method.