Liner.ai, no code end-to-end AI tools

This paragraph discusses the challenge for professors teaching machine learning (ML) or artificial intelligence (AI) in non-technical undergraduate courses, such as health sciences. It mentions how previously tools like Weka were used to simplify ML concepts but introduces a new tool, Liner AI, that facilitates end-to-end machine learning processes. The tool includes functionalities such as image recognition, classification, segmentation, and natural language processing (NLP). Additionally, it provides sample datasets, such as animals, plants, and traffic sign recognition. However, it highlights that Linux support is not yet available, with the current versions working on Windows and Mac.

This section explains how Liner AI can be used to train models, including a focus on image classification. The speaker shares their experience training a model using EfficientNet and mentions pre-trained models like MobileNet and ResNet. They describe the training process, including extracting features and how the tool replaces top layers with custom data. The training time on a MacBook Air is also discussed, along with performance metrics such as accuracy and recall. The paragraph highlights that while the accuracy was decent, the recall varied significantly, especially due to the small dataset size.

Here, the speaker delves into exporting and testing AI models using real-time external data. The exported models can be deployed in Python-based web applications, such as Flask or Django, or even mobile apps. The speaker notes that Liner AI produces compact model files and demonstrates testing the model with external images. They mention that the training process could be smoother with more training steps but generally find the export feature convenient, especially for creating mobile-ready models.

This paragraph focuses on training NLP models using Liner AI. The speaker describes working with distilled versions of models, such as DistilBERT, which allow for faster and smaller model processing. They mention training models with a dataset of up to 1100 entries, achieving an accuracy of 87% across multiple classes. The summary includes a brief comparison of the speaker's experience with a previous competition where they used different models for text classification.

In the final paragraph, the speaker reflects on Liner AI as a beginner-friendly tool for those new to AI, particularly because it requires no coding. The speaker compares it to other machine learning tools like Weka, highlighting the benefits of Liner AI for non-technical users. They briefly mention tools for more advanced users, such as XGBoost, LightGBM, and CatBoost, and suggest that these could be topics for future discussions.