Abstract:Automated Machine Learning (AutoML) has simplified complex ML processes such as data pre-processing, model selection, and hyper-parameter searching. However, traditional AutoML frameworks focus solely on discriminative tasks, often falling short in tackling AutoML for generative models. Additionally, these frameworks lack interpretability and user engagement during the training process, primarily due to the absence of human-centered design. It leads to a lack of transparency in final decision-making and limited user control, potentially reducing trust and adoption of AutoML methods. To address these limitations, we introduce UniAutoML, a human-centered AutoML framework that leverages Large Language Models (LLMs) to unify AutoML for both discriminative (e.g., Transformers and CNNs for classification or regression tasks) and generative tasks (e.g., fine-tuning diffusion models or LLMs). The human-centered design of UniAutoML innovatively features a conversational user interface (CUI) that facilitates natural language interactions, providing users with real-time guidance, feedback, and progress updates for better interpretability. This design enhances transparency and user control throughout the AutoML training process, allowing users to seamlessly break down or modify the model being trained. To mitigate potential risks associated with LLM generated content, UniAutoML incorporates a safety guardline that filters inputs and censors outputs. We evaluated UniAutoML's performance and usability through experiments on eight diverse datasets and user studies involving 25 participants, demonstrating that UniAutoML not only enhances performance but also improves user control and trust. Our human-centered design bridges the gap between AutoML capabilities and user understanding, making ML more accessible to a broader audience.

What problem does this paper attempt to address?

### Problems the paper attempts to solve This paper aims to solve several key problems existing in the current automated machine learning (AutoML) frameworks: 1. **Limitations in task types**: - Existing AutoML frameworks mainly focus on discriminative tasks, such as classification and regression, while generative tasks (such as fine - tuning of diffusion models and large language models) have not been fully explored and supported. This leads to deficiencies in handling generative tasks, especially for non - expert users, and the existing AutoML solutions seem not comprehensive enough. 2. **Lack of interpretability and user engagement**: - Traditional AutoML frameworks often lack transparency and user engagement. These frameworks usually operate in a "black - box" form, providing limited explanations and lacking real - time feedback, which reduces users' trust and hinders the wide adoption of AutoML tools. 3. **Limited user control ability**: - It is difficult for users to make modifications or optimizations during the model development and training process, which limits the users' control ability over the entire machine learning pipeline, especially when facing changing requirements. 4. **Security issues**: - There are potential security and ethical risks when using large language models (LLM), and effective mechanisms are required to filter inputs and review outputs to ensure the security and reliability of the content. ### Solutions To solve the above problems, the paper proposes **UniAutoML**, a user - centered AutoML framework with the following features: 1. **Unifying discriminative and generative tasks**: - UniAutoML not only supports traditional discriminative models (such as Transformer and CNN), but also extends to advanced generative models (such as diffusion models and large language models), and realizes automatic fine - tuning through natural language commands, reducing the usage threshold for non - expert users. 2. **Enhanced interpretability and user engagement**: - By introducing a conversational user interface (CUI), UniAutoML provides detailed AutoML output and progress information, and allows users to modify model parameters in real - time or terminate poorly - performing trials, thereby enhancing users' understanding and control ability. 3. **A wide library of pre - trained models**: - UniAutoML is integrated with HuggingFace and can access a large number of pre - trained models, significantly improving the flexibility and extensibility of the framework, and users can easily obtain the latest models suitable for their tasks. 4. **Security mechanism**: - A security protection mechanism is implemented to filter user inputs and review LLM outputs, ensuring the security and ethics of the content. ### Summary UniAutoML solves the limitations of existing AutoML frameworks by unifying discriminative and generative tasks, enhancing interpretability and user engagement, providing a wide library of pre - trained models, and implementing a security mechanism, making machine learning more accessible and intuitive for a wider range of user groups.

UniAutoML: A Human-Centered Framework for Unified Discriminative and Generative AutoML with Large Language Models

AutoM3L: An Automated Multimodal Machine Learning Framework with Large Language Models

UniGen: A Unified Framework for Textual Dataset Generation Using Large Language Models

AutoMMLab: Automatically Generating Deployable Models from Language Instructions for Computer Vision Tasks

Unified Language Model Pre-training for Natural Language Understanding and Generation

UniDM: A Unified Framework for Data Manipulation with Large Language Models

AutoML: From Methodology to Application

AdaptoML-UX: An Adaptive User-centered GUI-based AutoML Toolkit for Non-AI Experts and HCI Researchers

CO-AutoML: an Optimizable Automated Machine Learning System

AutoML: A Perspective where Industry Meets Academy

UniAudio: An Audio Foundation Model Toward Universal Audio Generation

UniAudio: Towards Universal Audio Generation with Large Language Models

Whither AutoML? Understanding the Role of Automation in Machine Learning Workflows

Taking Human out of Learning Applications: A Survey on Automated Machine Learning.

AutoML in The Wild: Obstacles, Workarounds, and Expectations

Uni3D-LLM: Unifying Point Cloud Perception, Generation and Editing with Large Language Models

Uni3DL: Unified Model for 3D and Language Understanding

AutoML-Agent: A Multi-Agent LLM Framework for Full-Pipeline AutoML

UnifiedMLLM: Enabling Unified Representation for Multi-modal Multi-tasks With Large Language Model