Magnetocardiography (MCG) enables passive detection of weak magnetic fields generated by the heart with high sensitivity, which can offer valuable information for diagnosing and treating heart conditions. Due to the limitations of the geomagnetic field and unknown magnetic interference, the MCG signals are often overwhelmed by high levels of magnetic noise. In this paper, we propose the design of a high-resolution and movable MCG system comprised of an active-passive coupling magnetic control (AP-CMC) system and a wearable multi-channel signal detection array. The system realizes the MCG measurement at the same time as the AP-CMC system eliminates interference in real time, i.e., simultaneous control and simultaneous measurement. Dynamic MCG signal measurements were successfully conducted, obtaining typical characteristic features of MCG signals. Our method shows promise in enhancing the accuracy and expanding the scope of MCG measurement applications, thereby offering valuable insights for the early diagnosis and precise localization of heart diseases.

Large Language models (LLMs) have become a research hotspot. To accelerate the inference of LLMs, storing computed caches in memory has become the standard technique. However, as the inference length increases, growing KV caches might lead to out-of-memory issues. Many existing methods address this issue through KV cache compression, primarily by preserving key tokens throughout all layers to reduce information loss. Most of them allocate a uniform budget size for each layer to retain. However, we observe that the minimum budget sizes needed to retain essential information vary across layers and models based on the perspectives of attention and hidden state output. Building on this observation, this paper proposes a simple yet effective KV cache compression method that leverages layer uncertainty to allocate budget size for each layer. Experimental results show that the proposed method can reduce memory usage of the KV caches to only ~20% when compared to full KV inference while achieving nearly lossless performance.

Enabling LLMs to handle lengthy context is currently a research hotspot. Most LLMs are built upon rotary position embedding (RoPE), a popular position encoding method. Therefore, a prominent path is to extrapolate the RoPE trained on comparably short texts to far longer texts. A heavy bunch of efforts have been dedicated to boosting the extrapolation via extending the formulations of the RoPE, however, few of them have attempted to showcase their inner workings comprehensively. In this paper, we are driven to offer a straightforward yet in-depth understanding of RoPE extensions from an attention perspective and on two benchmarking tasks. A broad array of experiments reveals several valuable findings: 1) Maintaining attention patterns to those at the pretrained length improves extrapolation; 2) Large attention uncertainty leads to retrieval errors; 3) Using longer continual pretraining lengths for RoPE extensions could reduce attention uncertainty and significantly enhance extrapolation.

The Magnetoencephalography (MEG) based on atomic magnetometer has high demand for the uniformity in the measurement area. Due to the door and holes of the magnetically shielding room (MSR), the distribution of residual magnetic field is non-linear, introducing much trouble into the magnetic field compensation. This study proposes a method for improving the uniformity of the residual magnetic field in the compact MSR by injecting different currents into the two coils in each pair of the outside compensation coils. The model and theoretical analysis method have been established, the feasibility has been verified both by the simulation and the experimental test. Results show that the maximum of the three components of the residual magnetic field in the direction of the three axes can be reduced by 18 times, 10.6 times and 8.1 times, respectively, when compared with the results without compensation, and 2.4 times, 2.3 times and 2.2 times, respectively, when compared with the results compensated with the same current. This study can provide certain help for achieving ultra-weak magnetic environment with high uniformity for the MEG measurementNote to Practitioners—In this article, we focus on the problem of compensating the residual magnetic field in the compact MSR with high efficiency and low cost, which remains the key factor that limits the promotion of the MEG. Method based on the compensation coils in series outside the MSR only deals with the linear or symmetric components, but has nothing to do with the non-linear and asymmetric one caused by the door. We propose a new method for compensating the residual magnetic field based on the coupling between the active and passive shielding, high compensation efficiency and uniformity can be achieved. In future work, we will aim at applying this method to create an ultra-weak magnetic field environment with high uniformity for the MEG measurement.

Ferromagnetic boundary coupling is one key factor that limits the control accuracy of the residual magnetic field in the magnetic shielding room. Usually the image method is utilized to suppress the coupling, but the tolerance for door slit and holes of the image method is poor. Here, we propose the self-shielding method for suppressing the ferromagnetic boundary coupling with higher coupling suppression ability and better tolerance for door slit and holes. Two sets of biplane coils for one axis are designed for achieving uniform magnetic field in the target area and nearzero magnetic field in the shielding area near the passive metal shielding layer simultaneously. We have determined the analysis method and design process of the self-shielding method, and verified its feasibility by experimental test. The proposed method can provide new ideas for improving the control accuracy of the residual magnetic field in the magnetic shielding room, ensuring the smooth progress of magnetoencephalography measurement.

We present AliMe Avatar, a Vtuber designed for live-streaming sales in the E-commerce field. To support the emerging live shopping mode, the core of our digitial avatar is to enable customers to understand products and encourage customers to purchase in a virtual broadcasting room. Based on computer graphics & vision, natural language processing, and speech recognition & synthesis, our AI avatar is able to offer three kinds of key capabilities: custom appearance, product broadcasting, and multi-modal interaction. Currently, it has been launched online in the Taobao app, broadcasts 700+ hours and serves hundreds of thousands of customers per day. In this paper, we mainly focus on the product broadcasting part, demonstrate the system, present the underlying techniques, and share our experience in dealing with live-streaming E-commerce.

Dongyu Zhang, Heting Zhang, Xikai Liu, Hongfei Lin, Feng Xia. Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP). 2019.