D. M. Li

Chengrui Gao, Haopu Shang, Ke Xue et al.

Machine learning has been adapted to help solve NP-hard combinatorial optimization problems. One prevalent way is learning to construct solutions by deep neural networks, which has been receiving more and more attention due to the high efficiency and less requirement for expert knowledge. However, many neural construction methods for Vehicle Routing Problems~(VRPs) focus on synthetic problem instances with specified node distributions and limited scales, leading to poor performance on real-world problems which usually involve complex and unknown node distributions together with large scales. To make neural VRP solvers more practical, we design an auxiliary policy that learns from the local transferable topological features, named local policy, and integrate it with a typical construction policy (which learns from the global information of VRP instances) to form an ensemble policy. With joint training, the aggregated policies perform cooperatively and complementarily to boost generalization. The experimental results on two well-known benchmarks, TSPLIB and CVRPLIB, of travelling salesman problem and capacitated VRP show that the ensemble policy significantly improves both cross-distribution and cross-scale generalization performance, and even performs well on real-world problems with several thousand nodes.

Lei Zhang, Dong Li, Olha Jurečková et al.

As machine learning and deep learning models become ubiquitous, it is inevitable that there will be attempts to exploit such models in various attack scenarios. For example, in a steganographic-based attack, information could be hidden in a learning model, which might then be used to distribute malware, or for other malicious purposes. In this research, we consider the steganographic capacity of several learning models. Specifically, we train a Multilayer Perceptron (MLP), Convolutional Neural Network (CNN), and Transformer model on a challenging malware classification problem. For each of the resulting models, we determine the number of low-order bits of the trained parameters that can be altered without significantly affecting the performance of the model. We find that the steganographic capacity of the learning models tested is surprisingly high, and that in each case, there is a clear threshold after which model performance rapidly degrades.

Dong Li, Divya Bhargavi, Vidya Sagar Ravipati

While recommender systems have significantly benefited from implicit feedback, they have often missed the nuances of multi-behavior interactions between users and items. Historically, these systems either amalgamated all behaviors, such as \textit{impression} (formerly \textit{view}), \textit{add-to-cart}, and \textit{buy}, under a singular 'interaction' label, or prioritized only the target behavior, often the \textit{buy} action, discarding valuable auxiliary signals. Although recent advancements tried addressing this simplification, they primarily gravitated towards optimizing the target behavior alone, battling with data scarcity. Additionally, they tended to bypass the nuanced hierarchy intrinsic to behaviors. To bridge these gaps, we introduce the \textbf{H}ierarchical \textbf{M}ulti-behavior \textbf{G}raph Attention \textbf{N}etwork (HMGN). This pioneering framework leverages attention mechanisms to discern information from both inter and intra-behaviors while employing a multi-task Hierarchical Bayesian Personalized Ranking (HBPR) for optimization. Recognizing the need for scalability, our approach integrates a specialized multi-behavior sub-graph sampling technique. Moreover, the adaptability of HMGN allows for the seamless inclusion of knowledge metadata and time-series data. Empirical results attest to our model's prowess, registering a notable performance boost of up to 64\% in NDCG@100 metrics over conventional graph neural network methods.

Yehui Tang, Yichun Yin, Yaoyuan Wang et al.

Sparse large language models (LLMs) with Mixture of Experts (MoE) and close to a trillion parameters are dominating the realm of most capable language models. However, the massive model scale poses significant challenges for the underlying software and hardware systems. In this paper, we aim to uncover a recipe to harness such scale on Ascend NPUs. The key goals are better usage of the computing resources under the dynamic sparse model structures and materializing the expected performance gain on the actual hardware. To select model configurations suitable for Ascend NPUs without repeatedly running the expensive experiments, we leverage simulation to compare the trade-off of various model hyperparameters. This study led to Pangu Ultra MoE, a sparse LLM with 718 billion parameters, and we conducted experiments on the model to verify the simulation results. On the system side, we dig into Expert Parallelism to optimize the communication between NPU devices to reduce the synchronization overhead. We also optimize the memory efficiency within the devices to further reduce the parameter and activation management overhead. In the end, we achieve an MFU of 30.0% when training Pangu Ultra MoE, with performance comparable to that of DeepSeek R1, on 6K Ascend NPUs, and demonstrate that the Ascend system is capable of harnessing all the training stages of the state-of-the-art language models. Extensive experiments indicate that our recipe can lead to efficient training of large-scale sparse language models with MoE. We also study the behaviors of such models for future reference.

Dong Li, Bo Ouyang, Duanpo Wu et al.

AGVs are driverless robotic vehicles that picks up and delivers materials. How to improve the efficiency while preventing deadlocks is the core issue in designing AGV systems. In this paper, we propose an approach to tackle this problem.The proposed approach includes a traditional AGV scheduling algorithm, which aims at solving deadlock problems, and an artificial neural network based component, which predict future tasks of the AGV system, and make decisions on whether to send an AGV to the predicted starting location of the upcoming task,so as to save the time of waiting for an AGV to go to there first when the upcoming task is created. Simulation results show that the proposed method significantly improves the efficiency as against traditional method, up to 20% to 30%.