Stephen Lau

Mingqi Li, Karan Aggarwal, Yong Xie et al.

As LLMs evolve, significant effort is spent on manually crafting prompts. While existing prompt optimization methods automate this process, they rely solely on learning from incorrect samples, leading to a sub-optimal performance. Additionally, an unexplored challenge in the literature is prompts effective for prior models may not perform well on newer versions or different languages. We propose the Learning from Contrastive Prompts (LCP) framework to address these gaps, enhancing both prompt optimization and adaptation. LCP employs contrastive learning to generate effective prompts by analyzing patterns in good and bad prompt examples. Our evaluation on the Big-Bench Hard dataset shows that LCP has a win rate of over 76% over existing methods in prompt optimization and demonstrates strong adaptability across different model versions, families, and languages. LCP offers a systematic approach to prompt engineering, reducing manual effort in deploying LLMs across varied contexts.

Stephen L. H. Lau, Edwin K. P. Chong

Single-pixel imaging (SPI) is a novel imaging technique whose working principle is based on the compressive sensing (CS) theory. In SPI, data is obtained through a series of compressive measurements and the corresponding image is reconstructed. Typically, the reconstruction algorithm such as basis pursuit relies on the sparsity assumption in images. However, recent advances in deep learning have found its uses in reconstructing CS images. Despite showing a promising result in simulations, it is often unclear how such an algorithm can be implemented in an actual SPI setup. In this paper, we demonstrate the use of deep learning on the reconstruction of SPI images in conjunction with block compressive sensing (BCS). We also proposed a novel reconstruction model based on convolutional neural networks that outperforms other competitive CS reconstruction algorithms. Besides, by incorporating BCS in our deep learning model, we were able to reconstruct images of any size above a certain smallest image size. In addition, we show that our model is capable of reconstructing images obtained from an SPI setup while being priorly trained on natural images, which can be vastly different from the SPI images. This opens up opportunity for the feasibility of pretrained deep learning models for CS reconstructions of images from various domain areas.

Jeffrey Eben, Aitzaz Ahmad, Stephen Lau

Despite advances in large language model (LLM)-based natural language interfaces for databases, scaling to enterprise-level data catalogs remains an under-explored challenge. Prior works addressing this challenge rely on domain-specific fine-tuning - complicating deployment - and fail to leverage important semantic context contained within database metadata. To address these limitations, we introduce a component-based retrieval architecture that decomposes database schemas and metadata into discrete semantic units, each separately indexed for targeted retrieval. Our approach prioritizes effective table identification while leveraging column-level information, ensuring the total number of retrieved tables remains within a manageable context budget. Experiments demonstrate that our method maintains high recall and accuracy, with our system outperforming baselines over massive databases with varying structure and available metadata. Our solution enables practical text-to-SQL systems deployable across diverse enterprise settings without specialized fine-tuning, addressing a critical scalability gap in natural language database interfaces.

Yong Xie, Karan Aggarwal, Aitzaz Ahmad et al.

We present a novel approach to automatically generate non-trivial task-specific synthetic datasets for hallucination detection. Our approach features a two-step generation-selection pipeline, using hallucination pattern guidance and a language style alignment during generation. Hallucination pattern guidance leverages the most important task-specific hallucination patterns while language style alignment aligns the style of the synthetic dataset with benchmark text. To obtain robust supervised detectors from synthetic datasets, we also adopt a data mixture strategy to improve performance robustness and generalization. Our results on three datasets show that our generated hallucination text is more closely aligned with non-hallucinated text versus baselines, to train hallucination detectors with better generalization. Our hallucination detectors trained on synthetic datasets outperform in-context-learning (ICL)-based detectors by a large margin of 32%. Our extensive experiments confirm the benefits of our approach with cross-task and cross-generator generalization. Our data-mixture-based training further improves the generalization and robustness of hallucination detection.

Stephen L. H. Lau, Edwin K. P. Chong, Xu Yang et al.

Automated pavement crack image segmentation is challenging because of inherent irregular patterns, lighting conditions, and noise in images. Conventional approaches require a substantial amount of feature engineering to differentiate crack regions from non-affected regions. In this paper, we propose a deep learning technique based on a convolutional neural network to perform segmentation tasks on pavement crack images. Our approach requires minimal feature engineering compared to other machine learning techniques. We propose a U-Net-based network architecture in which we replace the encoder with a pretrained ResNet-34 neural network. We use a "one-cycle" training schedule based on cyclical learning rates to speed up the convergence. Our method achieves an F1 score of 96% on the CFD dataset and 73% on the Crack500 dataset, outperforming other algorithms tested on these datasets. We perform ablation studies on various techniques that helped us get marginal performance boosts, i.e., the addition of spatial and channel squeeze and excitation (SCSE) modules, training with gradually increasing image sizes, and training various neural network layers with different learning rates.