$H^3$Fusion: Helpful, Harmless, Honest Fusion of Aligned LLMs

Alignment of pretrained LLMs using instruction-based datasets is critical for creating fine-tuned models that reflect human preference. A growing number of alignment-based fine-tuning algorithms and benchmarks emerged recently, fueling the efforts on effective alignments of pre-trained LLMs to ensure helpful, harmless, and honest answers from both open-source and closed-source LLMs. This paper tackles this problem by developing an alignment fusion approach, coined as $H^3$Fusion, with three unique characteristics. First, $H^3$Fusion ensembles multiple individually aligned LLMs to create a final fine-tuned alignment model with enhanced capabilities beyond those of individual models, delivering robust alignment through promoting helpful, harmless, honest fusion. Second, $H^3$Fusion leverages the mixture-of-experts (MoE) methodology in two steps. We first freeze the multi-head attention weights of each individual model while tuning the FFN layer during alignment fusion. Then we merge the aligned model weights with an expert router according to the type of input instruction and dynamically select a subset of experts that are best suited for producing the output response. Finally, we boost the performance of the resulting $H^3$3Fusion model by introducing gating loss and regularization terms. The former penalizes the selection errors of the expert-router, and the latter mediates the expert weights drifting during fine-tuning and dynamically adjusts the fusion behavior of the resulting model by canalizing the activations on the experts. Extensive evaluations on three benchmark datasets show that $H^3$3Fusion is more helpful, less harmful, and more honest from two aspects: it outperforms each individually aligned model by $11.37\%$, and it provides stronger robustness compared to the state-of-the-art LLM ensemble approaches by $13.77\%$. Code is available at github.com/sftekin/h3fusion.

Feature Extraction and Classification from Planetary Science Datasets enabled by Machine Learning

In this paper we present two examples of recent investigations that we have undertaken, applying Machine Learning (ML) neural networks (NN) to image datasets from outer planet missions to achieve feature recognition. Our first investigation was to recognize ice blocks (also known as rafts, plates, polygons) in the chaos regions of fractured ice on Europa. We used a transfer learning approach, adding and training new layers to an industry-standard Mask R-CNN (Region-based Convolutional Neural Network) to recognize labeled blocks in a training dataset. Subsequently, the updated model was tested against a new dataset, achieving 68% precision. In a different application, we applied the Mask R-CNN to recognize clouds on Titan, again through updated training followed by testing against new data, with a precision of 95% over 369 images. We evaluate the relative successes of our techniques and suggest how training and recognition could be further improved. The new approaches we have used for planetary datasets can further be applied to similar recognition tasks on other planets, including Earth. For imagery of outer planets in particular, the technique holds the possibility of greatly reducing the volume of returned data, via onboard identification of the most interesting image subsets, or by returning only differential data (images where changes have occurred) greatly enhancing the information content of the final data stream.