Abstract:Ensemble reasoning for the strengths of different LLM experts is critical to achieving consistent and satisfactory performance on diverse inputs across a wide range of tasks. However, existing LLM ensemble methods are either computationally intensive or incapable of leveraging complementary knowledge among LLM experts for various inputs. In this paper, we propose a Dynamic Ensemble Reasoning paradigm, called DER to integrate the strengths of multiple LLM experts conditioned on dynamic inputs. Specifically, we model the LLM ensemble reasoning problem as a Markov Decision Process (MDP), wherein an agent sequentially takes inputs to request knowledge from an LLM candidate and passes the output to a subsequent LLM candidate. Moreover, we devise a reward function to train a DER-Agent to dynamically select an optimal answering route given the input questions, aiming to achieve the highest performance with as few computational resources as possible. Last, to fully transfer the expert knowledge from the prior LLMs, we develop a Knowledge Transfer Prompt (KTP) that enables the subsequent LLM candidates to transfer complementary knowledge effectively. Experiments demonstrate that our method uses fewer computational resources to achieve better performance compared to state-of-the-art baselines.
Abstract:Multimodal LLMs (MLLMs) equip language models with visual capabilities by aligning vision encoders with language models. Existing methods to enhance the visual perception of MLLMs often involve designing more powerful vision encoders, which requires exploring a vast design space and re-aligning each potential encoder with the language model, resulting in prohibitively high training costs. In this paper, we introduce VisionFuse, a novel integration framework that efficiently utilizes multiple vision encoders from off-the-shelf MLLMs to enhance visual perception without requiring additional training. Our approach is motivated by the observation that different MLLMs tend to focus on distinct regions given the same query and image. Moreover, we find that the feature distributions of vision encoders within an MLLM family, a group of MLLMs sharing the same pretrained LLM, are highly aligned. Building on these insights, VisionFuse enriches the visual context by concatenating the tokens generated by the vision encoders of selected MLLMs within a family. By merging the parameters of language models from these MLLMs, VisionFuse allows a single language model to align with various vision encoders, significantly reducing deployment overhead. We conduct comprehensive evaluations across multiple multimodal benchmarks using various MLLM combinations, demonstrating substantial improvements in multimodal tasks. Notably, when integrating MiniGemini-8B and SLIME-8B, VisionFuse achieves an average performance increase of over 4%.
Abstract:Drone Visual Active Tracking aims to autonomously follow a target object by controlling the motion system based on visual observations, providing a more practical solution for effective tracking in dynamic environments. However, accurate Drone Visual Active Tracking using reinforcement learning remains challenging due to the absence of a unified benchmark, the complexity of open-world environments with frequent interference, and the diverse motion behavior of dynamic targets. To address these issues, we propose a unified cross-scene cross-domain benchmark for open-world drone active tracking called DAT. The DAT benchmark provides 24 visually complex environments to assess the algorithms' cross-scene and cross-domain generalization abilities, and high-fidelity modeling of realistic robot dynamics. Additionally, we propose a reinforcement learning-based drone tracking method called R-VAT, which aims to improve the performance of drone tracking targets in complex scenarios. Specifically, inspired by curriculum learning, we introduce a Curriculum-Based Training strategy that progressively enhances the agent tracking performance in vast environments with complex interference. We design a goal-centered reward function to provide precise feedback to the drone agent, preventing targets farther from the center of view from receiving higher rewards than closer ones. This allows the drone to adapt to the diverse motion behavior of open-world targets. Experiments demonstrate that the R-VAT has about 400% improvement over the SOTA method in terms of the cumulative reward metric.