Visual instruction tuning is an essential approach to improving the zero-shot generalization capability of Multi-modal Large Language Models (MLLMs). A surge of visual instruction datasets with various focuses and characteristics have been proposed recently, enabling MLLMs to achieve surprising results on evaluation benchmarks. To develop more capable MLLMs, in this paper, we aim to investigate a more fundamental question: ``what makes for good visual instructions?''. By conducting a comprehensive empirical study, we find that instructions focused on complex visual reasoning tasks are particularly effective in improving the performance of MLLMs on evaluation benchmarks. Building upon this finding, we design a systematic approach to automatically creating high-quality complex visual reasoning instructions. Our approach employs a synthesis-complication-reformulation paradigm, leveraging multiple stages to gradually increase the complexity of the instructions while guaranteeing quality. Based on this approach, we create the synthetic visual reasoning instruction dataset consisting of 32K examples, namely ComVint, and fine-tune four MLLMs on it. Experimental results demonstrate that our dataset consistently enhances the performance of all the compared MLLMs, e.g., improving the performance of MiniGPT-4 and BLIP-2 on MME-Cognition by 32.6% and 28.8%, respectively. Our code and data are publicly available at the link: https://github.com/RUCAIBox/ComVint.

As an effective tool for eliciting the power of Large Language Models (LLMs), prompting has recently demonstrated unprecedented abilities across a variety of complex tasks. To further improve the performance, prompt ensemble has attracted substantial interest for tackling the hallucination and instability of LLMs. However, existing methods usually adopt a two-stage paradigm, which requires a pre-prepared set of prompts with substantial manual effort, and is unable to perform directed optimization for different weak learners. In this paper, we propose a simple, universal, and automatic method named PREFER (Pompt Ensemble learning via Feedback-Reflect-Refine) to address the stated limitations. Specifically, given the fact that weak learners are supposed to focus on hard examples during boosting, PREFER builds a feedback mechanism for reflecting on the inadequacies of existing weak learners. Based on this, the LLM is required to automatically synthesize new prompts for iterative refinement. Moreover, to enhance stability of the prompt effect evaluation, we propose a novel prompt bagging method involving forward and backward thinking, which is superior to majority voting and is beneficial for both feedback and weight calculation in boosting. Extensive experiments demonstrate that our PREFER achieves state-of-the-art performance in multiple types of tasks by a significant margin. We have made our code publicly available.

AUC is a common metric for evaluating the performance of a classifier. However, most classifiers are trained with cross entropy, and it does not optimize the AUC metric directly, which leaves a gap between the training and evaluation stage. In this paper, we propose the PDAOM loss, a Personalized and Differentiable AUC Optimization method with Maximum violation, which can be directly applied when training a binary classifier and optimized with gradient-based methods. Specifically, we construct the pairwise exponential loss with difficult pair of positive and negative samples within sub-batches grouped by user ID, aiming to guide the classifier to pay attention to the relation between hard-distinguished pairs of opposite samples from the perspective of independent users. Compared to the origin form of pairwise exponential loss, the proposed PDAOM loss not only improves the AUC and GAUC metrics in the offline evaluation, but also reduces the computation complexity of the training objective. Furthermore, online evaluation of the PDAOM loss on the 'Guess What You Like' feed recommendation application in Meituan manifests 1.40% increase in click count and 0.65% increase in order count compared to the baseline model, which is a significant improvement in this well-developed online life service recommendation system.