International audience

LLM-driven multi-agent collaboration (MAC) systems have demonstrated impressive capabilities in automatic software development at the function level. However, their heavy reliance on human design limits their adaptability to the diverse demands of real-world software development. To address this limitation, we introduce EvoMAC, a novel self-evolving paradigm for MAC networks. Inspired by traditional neural network training, EvoMAC obtains text-based environmental feedback by verifying the MAC network's output against a target proxy and leverages a novel textual backpropagation to update the network. To extend coding capabilities beyond function-level tasks to more challenging software-level development, we further propose RSD-Bench, a requirement-oriented software development benchmark, which features complex and diverse software requirements along with automatic evaluation of requirement correctness. Our experiments show that: i) The automatic requirement-aware evaluation in RSD-Bench closely aligns with human evaluations, validating its reliability as a software-level coding benchmark. ii) EvoMAC outperforms previous SOTA methods on both the software-level RSD-Bench and the function-level HumanEval benchmarks, reflecting its superior coding capabilities.

ABSTRACT With the constant dynamics of temporal dependence and spatial correlation, the interaction between them has become intricate. Existing work attempts to model precise temporal dependency and spatial correlation to make their interactions more accurate but ignores the importance of understanding how the two interact with each other. Thus, this article mines deeper into their interaction mechanism and proposes a new traffic prediction model called traffic flow prediction model based on spatial–temporal identity (TPST). It provides a new way named the spatial–temporal identity mechanism to model spatial–temporal interactions, which convert complex temporal dependence and spatial correlation into their identity information. Meanwhile, in order to improve spatial–temporal interaction resolution of the model, the method utilizes the down‐sampling cross‐convolution technique to contain more spatial–temporal history information and parses spatial–temporal interactions at different granularity. Experiments conducted with four real traffic flow datasets show that TPST consistently outperforms the other seven benchmark models, providing higher prediction accuracy with lower computational cost.

Despite their immense success as a model of macaque visual cortex, deep convolutional neural networks (CNNs) have struggled to predict activity in visual cortex of the mouse, which is thought to be strongly dependent on the animal’s behavioral state. Furthermore, most computational models focus on predicting neural responses to static images presented under head fixation, which are dramatically different from the dynamic, continuous visual stimuli that arise during movement in the real world. Consequently, it is still unknown how natural visual input and different behavioral variables may integrate over time to generate responses in primary visual cortex (V1). To address this, we introduce a multimodal recurrent neural network that integrates gaze-contingent visual input with behavioral and temporal dynamics to explain V1 activity in freely moving mice. We show that the model achieves state-of-the-art predictions of V1 activity during free exploration and demonstrate the importance of each component in an extensive ablation study. Analyzing our model using maximally activating stimuli and saliency maps, we reveal new insights into cortical function, including the prevalence of mixed selectivity for behavioral variables in mouse V1. In summary, our model offers a comprehensive deep-learning framework for exploring the computational principles underlying V1 neurons in freely-moving animals engaged in natural behavior.

Deep learning has been successful in various domains including image recognition, speech recognition and natural language processing. However, the research on its application in graph mining is still in an early stage. Here we present Model R, a neural network model created to provide a deep learning approach to link weight prediction problem. This model extracts knowledge of nodes from known links' weights and uses this knowledge to predict unknown links' weights. We demonstrate the power of Model R through experiments and compare it with stochastic block model and its derivatives. Model R shows that deep learning can be successfully applied to link weight prediction and it outperforms stochastic block model and its derivatives by up to 73% in terms of prediction accuracy. We anticipate this new approach to provide effective solutions to more graph mining tasks.