Large Language Models (LLMs) are known to hallucinate and generate non-factual outputs which can undermine user trust. Traditional methods to directly mitigate hallucinations, such as representation editing and contrastive decoding, often require additional training data and involve high implementation complexity. While ensemble-based approaches harness multiple LLMs to tap into the "wisdom of crowds", these methods overlook uncertainties in individual model responses. Recent studies reveal that uncertainty estimation can enable LLMs to self-assess the likelihood of generating hallucinations. In this work, we focus on factoid question answering (QA) and observe that LLMs accuracy and self-assessment capabilities vary widely with different models excelling in different scenarios. Leveraging this insight, we propose Uncertainty-Aware Fusion (UAF), an ensemble framework to reduces hallucinations by strategically combining multiple LLM based on their accuracy and self-assessment abilities. Empirical results on several public benchmark datasets show that UAF outperforms state-of-the-art hallucination mitigation methods by $8\%$ in factual accuracy, while either narrowing or surpassing the performance gap with GPT-4.

This article proposes a Universal Activation Function (UAF) that achieves near optimal performance in quantification, classification, and reinforcement learning (RL) problems. For any given problem, the optimization algorithms are able to evolve the UAF to a suitable activation function by tuning the UAF's parameters. For the CIFAR-10 classification and VGG-8, the UAF converges to the Mish like activation function, which has near optimal performance $F_{1} = 0.9017\pm0.0040$ when compared to other activation functions. For the quantification of simulated 9-gas mixtures in 30 dB signal-to-noise ratio (SNR) environments, the UAF converges to the identity function, which has near optimal root mean square error of $0.4888 \pm 0.0032$ $\mu M$. In the BipedalWalker-v2 RL dataset, the UAF achieves the 250 reward in $961 \pm 193$ epochs, which proves that the UAF converges in the lowest number of epochs. Furthermore, the UAF converges to a new activation function in the BipedalWalker-v2 RL dataset.

Missing data are a concern in many real world data sets and imputation methods are often needed to estimate the values of missing data, but data sets with excessive missingness and high dimensionality challenge most approaches to imputation. Here we show that appropriate feature selection can be an effective preprocessing step for imputation, allowing for more accurate imputation and subsequent model predictions. The key feature of this preprocessing is that it incorporates uncertainty: by accounting for uncertainty due to missingness when selecting features we can reduce the degree of missingness while also limiting the number of uninformative features being used to make predictive models. We introduce a method to perform uncertainty-aware feature selection (UAFS), provide a theoretical motivation, and test UAFS on both real and synthetic problems, demonstrating that across a variety of data sets and levels of missingness we can improve the accuracy of imputations. Improved imputation due to UAFS also results in improved prediction accuracy when performing supervised learning using these imputed data sets. Our UAFS method is general and can be fruitfully coupled with a variety of imputation methods.