FAIR Enough: How Can We Develop and Assess a FAIR-Compliant Dataset for Large Language Models' Training?

The rapid evolution of Large Language Models (LLMs) underscores the critical importance of ethical considerations and data integrity in AI development, emphasizing the role of FAIR (Findable, Accessible, Interoperable, Reusable) data principles. While these principles have long been a cornerstone of ethical data stewardship, their application in LLM training data is less prevalent, an issue our research aims to address. Our study begins with a review of existing literature, highlighting the significance of FAIR principles in data management for model training. Building on this foundation, we introduce a novel framework that incorporates FAIR principles into the LLM training process. A key aspect of this approach is a comprehensive checklist, designed to assist researchers and developers in consistently applying FAIR data principles throughout the model development lifecycle. The practicality and effectiveness of our framework are demonstrated through a case study that involves creating a FAIR-compliant dataset to detect and reduce biases. This case study not only validates the usefulness of our framework but also establishes new benchmarks for more equitable, transparent, and ethical practices in LLM training. We offer this framework to the community as a means to promote technologically advanced, ethically sound, and socially responsible AI models.

DMC4ML: Data Movement Complexity for Machine Learning

The greatest demand for today's computing is machine learning. This paper analyzes three machine learning algorithms: transformers, spatial convolution, and FFT. The analysis is novel in three aspects. First, it measures the cost of memory access on an abstract memory hierarchy, instead of traditional time or space complexity. Second, the analysis is asymptotic and identifies the primary sources of the memory cost. Finally, the result is symbolic, which can be used to select algorithmic parameters such as the group size in grouped query attention for any dimension size and number of heads and the batch size for batched convolution for any image size and kernel size.

Scalable CP Decomposition for Tensor Learning using GPU Tensor Cores

CP decomposition is a powerful tool for data science, especially gene analysis, deep learning, and quantum computation. However, the application of tensor decomposition is largely hindered by the exponential increment of the computational complexity and storage consumption with the size of tensors. While the data in our real world is usually presented as trillion- or even exascale-scale tensors, existing work can only support billion-scale scale tensors. In our work, we propose the Exascale-Tensor to mitigate the significant gap. Specifically, we propose a compression-based tensor decomposition framework, namely the exascale-tensor, to support exascale tensor decomposition. Then, we carefully analyze the inherent parallelism and propose a bag of strategies to improve computational efficiency. Last, we conduct experiments to decompose tensors ranging from million-scale to trillion-scale for evaluation. Compared to the baselines, the exascale-tensor supports 8,000x larger tensors and a speedup up to 6.95x. We also apply our method to two real-world applications, including gene analysis and tensor layer neural networks, of which the numeric results demonstrate the scalability and effectiveness of our method.

Bidirectional Graph GAN: Representing Brain Structure-Function Connections for Alzheimer's Disease

The relationship between brain structure and function is critical for revealing the pathogenesis of brain disease, including Alzheimer's disease (AD). However, it is a great challenge to map brain structure-function connections due to various reasons. In this work, a bidirectional graph generative adversarial networks (BGGAN) is proposed to represent brain structure-function connections. Specifically, by designing a module incorporating inner graph convolution network (InnerGCN), the generators of BGGAN can employ features of direct and indirect brain regions to learn the mapping function between structural domain and functional domain. Besides, a new module named Balancer is designed to counterpoise the optimization between generators and discriminators. By introducing the Balancer into BGGAN, both the structural generator and functional generator can not only alleviate the issue of mode collapse but also learn complementarity of structural and functional features. Experimental results using ADNI datasets show that the both the generated structure connections and generated function connections can improve the identification accuracy of AD. More importantly, based the proposed model, it is found that the relationship between brain structure and function is not a complete one-to-one correspondence. Brain structure is the basis of brain function. The strong structural connections are almost accompanied by strong functional connections.

NBIAS: A Natural Language Processing Framework for Bias Identification in Text

Bias in textual data can lead to skewed interpretations and outcomes when the data is used. These biases could perpetuate stereotypes, discrimination, or other forms of unfair treatment. An algorithm trained on biased data ends up making decisions that disproportionately impact a certain group of people. Therefore, it is crucial to detect and remove these biases to ensure the fair and ethical use of data. To this end, we develop a comprehensive and robust framework \textsc{Nbias} that consists of a data layer, corpus contruction, model development layer and an evaluation layer. The dataset is constructed by collecting diverse data from various fields, including social media, healthcare, and job hiring portals. As such, we applied a transformer-based token classification model that is able to identify bias words/ phrases through a unique named entity. In the assessment procedure, we incorporate a blend of quantitative and qualitative evaluations to gauge the effectiveness of our models. We achieve accuracy improvements ranging from 1% to 8% compared to baselines. We are also able to generate a robust understanding of the model functioning, capturing not only numerical data but also the quality and intricacies of its performance. The proposed approach is applicable to a variety of biases and contributes to the fair and ethical use of textual data.

Mitigating Bias in Conversations: A Hate Speech Classifier and Debiaser with Prompts

Discriminatory language and biases are often present in hate speech during conversations, which usually lead to negative impacts on targeted groups such as those based on race, gender, and religion. To tackle this issue, we propose an approach that involves a two-step process: first, detecting hate speech using a classifier, and then utilizing a debiasing component that generates less biased or unbiased alternatives through prompts. We evaluated our approach on a benchmark dataset and observed reduction in negativity due to hate speech comments. The proposed method contributes to the ongoing efforts to reduce biases in online discourse and promote a more inclusive and fair environment for communication.

Dbias: Detecting biases and ensuring Fairness in news articles

Because of the increasing use of data-centric systems and algorithms in machine learning, the topic of fairness is receiving a lot of attention in the academic and broader literature. This paper introduces Dbias (https://pypi.org/project/Dbias/), an open-source Python package for ensuring fairness in news articles. Dbias can take any text to determine if it is biased. Then, it detects biased words in the text, masks them, and suggests a set of sentences with new words that are bias-free or at least less biased. We conduct extensive experiments to assess the performance of Dbias. To see how well our approach works, we compare it to the existing fairness models. We also test the individual components of Dbias to see how effective they are. The experimental results show that Dbias outperforms all the baselines in terms of accuracy and fairness. We make this package (Dbias) as publicly available for the developers and practitioners to mitigate biases in textual data (such as news articles), as well as to encourage extension of this work.

Active Learning on a Programmable Photonic Quantum Processor

Training a quantum machine learning model generally requires a large labeled dataset, which incurs high labeling and computational costs. To reduce such costs, a selective training strategy, called active learning (AL), chooses only a subset of the original dataset to learn while maintaining the trained model's performance. Here, we design and implement two AL-enpowered variational quantum classifiers, to investigate the potential applications and effectiveness of AL in quantum machine learning. Firstly, we build a programmable free-space photonic quantum processor, which enables the programmed implementation of various hybrid quantum-classical computing algorithms. Then, we code the designed variational quantum classifier with AL into the quantum processor, and execute comparative tests for the classifiers with and without the AL strategy. The results validate the great advantage of AL in quantum machine learning, as it saves at most $85\%$ labeling efforts and $91.6\%$ percent computational efforts compared to the training without AL on a data classification task. Our results inspire AL's further applications in large-scale quantum machine learning to drastically reduce training data and speed up training, underpinning the exploration of practical quantum advantages in quantum physics or real-world applications.

Parameter-Parallel Distributed Variational Quantum Algorithm

Variational quantum algorithms (VQAs) have emerged as a promising near-term technique to explore practical quantum advantage on noisy intermediate-scale quantum (NISQ) devices. However, the inefficient parameter training process due to the incompatibility with backpropagation and the cost of a large number of measurements, posing a great challenge to the large-scale development of VQAs. Here, we propose a parameter-parallel distributed variational quantum algorithm (PPD-VQA), to accelerate the training process by parameter-parallel training with multiple quantum processors. To maintain the high performance of PPD-VQA in the realistic noise scenarios, a alternate training strategy is proposed to alleviate the acceleration attenuation caused by noise differences among multiple quantum processors, which is an unavoidable common problem of distributed VQA. Besides, the gradient compression is also employed to overcome the potential communication bottlenecks. The achieved results suggest that the PPD-VQA could provide a practical solution for coordinating multiple quantum processors to handle large-scale real-word applications.