Recent research has increasingly focused on reconciling the reasoning capabilities of System 2 with the efficiency of System 1. While existing training-based and prompt-based approaches face significant challenges in terms of efficiency and stability, model merging emerges as a promising strategy to integrate the diverse capabilities of different Large Language Models (LLMs) into a unified model. However, conventional model merging methods often assume uniform importance across layers, overlooking the functional heterogeneity inherent in neural components. To address this limitation, we propose \textbf{A}ctivation-Guided \textbf{C}onsensus \textbf{M}erging (\textbf{ACM}), a plug-and-play merging framework that determines layer-specific merging coefficients based on mutual information between activations of pre-trained and fine-tuned models. ACM effectively preserves task-specific capabilities without requiring gradient computations or additional training. Extensive experiments on Long-to-Short (L2S) and general merging tasks demonstrate that ACM consistently outperforms all baseline methods. For instance, in the case of Qwen-7B models, TIES-Merging equipped with ACM achieves a \textbf{55.3\%} reduction in response length while simultaneously improving reasoning accuracy by \textbf{1.3} points.

Multi-label active learning is a crucial yet challenging area in contemporary machine learning, often complicated by a large and sparse label space.

Anyone who has tried to swat a fly has likely been frustrated by its remarkable agility.This ability stems from its visual neural perception system, particularly the collision-selective neurons within its small brain.For autonomous robots operating in complex and unfamiliar environments, achieving similar agility is highly desirable but often constrained by the trade-off between computational cost and performance.In this context, insect-inspired intelligence offers a parsimonious route to low-power, computationally efficient frameworks.In this paper, we propose an attention-driven visuomotor control strategy inspired by a specific class of fly visual projection neurons-the lobula plate/lobula column type-2 (LPLC2)-and their associated escape behaviors.To our knowledge, this represents the first embodiment of an LPLC2 neural model in the embedded vision of a physical mobile robot, enabling collision perception and reactive evasion.The model was simplified and optimized at 70KB in memory to suit the computational constraints of a vision-based micro robot, the Colias, while preserving key neural perception mechanisms.We further incorporated multi-attention mechanisms to emulate the distributed nature of LPLC2 responses, allowing the robot to detect and react to approaching targets both rapidly and selectively.We systematically evaluated the proposed method against a state-of-the-art locust-inspired collision detection model.Results showed that the fly-inspired visuomotor model achieved comparable robustness, at success rate of 96.1% in collision detection while producing more adaptive and elegant evasive maneuvers.Beyond demonstrating an effective collision-avoidance strategy, this work highlights the potential of fly-inspired neural models for advancing research into collective behaviors in insect intelligence.

Lane keeping in autonomous driving systems requires scenario-specific weight tuning for different objectives. We formulate lane-keeping as a constrained reinforcement learning problem, where weight coefficients are automatically learned along with the policy, eliminating the need for scenario-specific tuning. Empirically, our approach outperforms traditional RL in efficiency and reliability. Additionally, real-world demonstrations validate its practical value for real-world autonomous driving.

Data augmentation is known to contribute significantly to the robustness of machine learning models. In most instances, data augmentation is utilized during the training phase. Test-Time Augmentation (TTA) is a technique that instead leverages these data augmentations during the testing phase to achieve robust predictions. More precisely, TTA averages the predictions of multiple data augmentations of an instance to produce a final prediction. Although the effectiveness of TTA has been empirically reported, it can be expected that the predictive performance achieved will depend on the set of data augmentation methods used during testing. In particular, the data augmentation methods applied should make different contributions to performance. That is, it is anticipated that there may be differing degrees of contribution in the set of data augmentation methods used for TTA, and these could have a negative impact on prediction performance. In this study, we consider a weighted version of the TTA based on the contribution of each data augmentation. Some variants of TTA can be regarded as considering the problem of determining the appropriate weighting. We demonstrate that the determination of the coefficients of this weighted TTA can be formalized in a variational Bayesian framework. We also show that optimizing the weights to maximize the marginal log-likelihood suppresses candidates of unwanted data augmentations at the test phase.

Generating counterfactual explanations is one of the most effective approaches for uncovering the inner workings of black-box neural network models and building user trust. While remarkable strides have been made in generative modeling using diffusion models in domains like vision, their utility in generating counterfactual explanations in structured modalities remains unexplored. In this paper, we introduce Structured Counterfactual Diffuser or SCD, the first plug-and-play framework leveraging diffusion for generating counterfactual explanations in structured data. SCD learns the underlying data distribution via a diffusion model which is then guided at test time to generate counterfactuals for any arbitrary black-box model, input, and desired prediction. Our experiments show that our counterfactuals not only exhibit high plausibility compared to the existing state-of-the-art but also show significantly better proximity and diversity.

In the tasks of multi-robot collaborative area search, we propose the unified approach for simultaneous mapping for sensing more targets (exploration) while searching and locating the targets (coverage). Specifically, we implement a hierarchical multi-agent reinforcement learning algorithm to decouple task planning from task execution. The role concept is integrated into the upper-level task planning for role selection, which enables robots to learn the role based on the state status from the upper-view. Besides, an intelligent role switching mechanism enables the role selection module to function between two timesteps, promoting both exploration and coverage interchangeably. Then the primitive policy learns how to plan based on their assigned roles and local observation for sub-task execution. The well-designed experiments show the scalability and generalization of our method compared with state-of-the-art approaches in the scenes with varying complexity and number of robots.

Maintaining both path-tracking accuracy and yaw stability of distributed drive electric vehicles (DDEVs) under various driving conditions presents a significant challenge in the field of vehicle control. To address this limitation, a coordinated control strategy that integrates adaptive model predictive control (AMPC) path-tracking control and direct yaw moment control (DYC) is proposed for DDEVs. The proposed strategy, inspired by a hierarchical framework, is coordinated by the upper layer of path-tracking control and the lower layer of direct yaw moment control. Based on the linear time-varying model predictive control (LTV MPC) algorithm, the effects of prediction horizon and weight coefficients on the path-tracking accuracy and yaw stability of the vehicle are compared and analyzed first. According to the aforementioned analysis, an AMPC path-tracking controller with variable prediction horizon and weight coefficients is designed considering the vehicle speed's variation in the upper layer. The lower layer involves DYC based on the linear quadratic regulator (LQR) technique. Specifically, the intervention rule of DYC is determined by the threshold of the yaw rate error and the phase diagram of the sideslip angle. Extensive simulation experiments are conducted to evaluate the proposed coordinated control strategy under different driving conditions. The results show that, under variable speed and low adhesion conditions, the vehicle's yaw stability and path-tracking accuracy have been improved by 21.58\% and 14.43\%, respectively, compared to AMPC. Similarly, under high speed and low adhesion conditions, the vehicle's yaw stability and path-tracking accuracy have been improved by 44.30\% and 14.25\%, respectively, compared to the coordination of LTV MPC and DYC. The results indicate that the proposed adaptive path-tracking controller is effective across different speeds.

Multitask deep learning has been applied to patient outcome prediction from text, taking clinical notes as input and training deep neural networks with a joint loss function of multiple tasks. However, the joint training scheme of multitask learning suffers from inter-task interference, and diagnosis prediction among the multiple tasks has the generalizability issue due to rare diseases or unseen diagnoses. To solve these challenges, we propose a hypernetwork-based approach that generates task-conditioned parameters and coefficients of multitask prediction heads to learn task-specific prediction and balance the multitask learning. We also incorporate semantic task information to improves the generalizability of our task-conditioned multitask model. Experiments on early and discharge notes extracted from the real-world MIMIC database show our method can achieve better performance on multitask patient outcome prediction than strong baselines in most cases. Besides, our method can effectively handle the scenario with limited information and improve zero-shot prediction on unseen diagnosis categories.