Large Language Models’ safety-aligned behaviors, such as refusing harmful queries, can be represented by linear directions in activation space. Previous research modeled safety behavior with a single direction, limiting mechanistic understanding to an isolated safety feature. In this work, we discover that safety-aligned behavior is jointly controlled by multi-dimensional directions. Namely, we study the vector space of representation shifts during safety fine-tuning on Llama 3 8B for refusing jailbreaks. By studying orthogonal directions in the space, we first find that a dominant direction governs the model’s refusal behavior, while multiple smaller directions represent distinct and interpretable features like hypothetical narrative and role-playing. We then measure how different directions promote or suppress the dominant direction, showing the important role of secondary directions in shaping the model’s refusal representation. Finally, we demonstrate that removing certain trigger tokens in harmful queries can mitigate these directions to bypass the learned safety capability, providing new insights on understanding safety alignment vulnerability from a multi-dimensional perspective.

In recent years, nuclear magnetic resonance gyroscope (NMRG) sensor has attracted significant attention due to its potential for miniaturization. The uniform magnetic field coil is the main component for manipulating atomic states, and its uniformity performance affects the performance of NMRG sensor. But as the volume of sensor decreases, the uniformity of the bi-planar coil also declines. In order to solve the problem of decreased uniformity caused by volume reduction, this article proposes an optimized target field method (TFM) based on group search optimizer (GSO) for high uniform bi-planar coil design. In order to select the optimal coil design parameters, a comparison of optimized TFM results under different GSO parameters has been conducted. To demonstrate the effectiveness of the proposed method, this article conducted magnetic field simulations on the designed coil and compared it with the traditional TFM. The results show that the uniformity performance of the proposed method reaches 7.01×10-4, and improving by 52% compared to the traditional TFM. This article provides a novel design method for the high uniform bi-planar coil of miniaturized NMRG sensor.

The real-time generation of editable mesh animations in XR applications has been a focal point of research in the XR field. However, easily controlling the generated editable meshes remains a significant challenge. Existing methods often suffer from slow generation speeds and suboptimal results, failing to accurately simulate target objects' complex details and shapes, which does not meet user expectations. Additionally, the final generated meshes typically require manual user adjustments, and it is difficult to generate multiple target models simultaneously. To overcome these limitations, a universal control scheme for particles based on the sampling features of the target is proposed. It introduces a spatially adaptive control algorithm for particle coupling by adjusting the magnitude of control forces based on the spatial features of model sampling, thereby eliminating the need for parameter dependency and enabling the control of multiple types of models within the same scene. We further introduce boundary correction techniques to improve the precision in generating target shapes while reducing particle splashing. Moreover, a distance-adaptive particle fragmentation mechanism prevents unnecessary particle accumulation. Experimental results demonstrate that the method has better performance in controlling complex structures and generating multiple targets at the same time compared to existing methods. It enhances control accuracy for complex structures and targets under the condition of sparse model sampling. It also consistently delivers outstanding results while maintaining high stability and efficiency. Ultimately, we were able to create a set of smooth editable meshes and developed a solution for integrating this algorithm into VR and AR animation applications.

Summary We proposed a novel target‐driven fluid simulation method based on the weighted control model derived from the spatial geometric features of the target shape. First, the spatial geometric characteristics of the target model are taken into account to set the color field weights of control particles. This enabled the full expression of geometric characteristics of the target model, and improve the shape accuracy of controlled fluid. Then, the fluid is controlled to form the target shape under driving constraints, wherein we proposed a new adaptive constraint mechanism that enables efficient target shape generation. Finally, a new density constraint between the control particles and the controlled fluid particles is proposed to ensure the incompressibility of fluid during control. Compared to the state‐of‐the‐art target‐driven fluid control methods, our method achieves higher precision fluid control with higher efficiency.

Though great progress has been made for human-machine conversation, current dialogue system is still in its infancy: it usually converses passively and utters words more as a matter of response, rather than on its own initiatives. In this paper, we take a radical step towards building a human-like conversational agent: endowing it with the ability of proactively leading the conversation (introducing a new topic or maintaining the current topic). To facilitate the development of such conversation systems, we create a new dataset named Konv where one acts as a conversation leader and the other acts as the follower. The leader is provided with a knowledge graph and asked to sequentially change the discussion topics, following the given conversation goal, and meanwhile keep the dialogue as natural and engaging as possible. Konv enables a very challenging task as the model needs to both understand dialogue and plan over the given knowledge graph. We establish baseline results on this dataset (about 270K utterances and 30k dialogues) using several state-of-the-art models. Experimental results show that dialogue models that plan over the knowledge graph can make full use of related knowledge to generate more diverse multi-turn conversations. The baseline systems along with the dataset are publicly available.

Marine current energy conversion systems are progressively attracting more attention due to its efficaciousness in providing a predictable source of energy with high energy density. This paper considers the critical issues such as the rejection of lumped disturbances, system uncertainties in the internal dynamics, and unknown external forces in the marine current energy conversion system (MCECS). An active disturbance rejection control (ADRC) strategy is applied to realize the maximum power extraction for a direct-driven permanent magnet synchronous generator (PMSG) based marine current turbine (MCT). The ADRC is used to realize the real-time estimation of system interference and interference compensation in order to improve the system ability to suppress errors and disturbances. Finally, this strategy is successfully tested through simulation, the results for which show that the control strategy is effective compared with conventional control.

Xiangyang Zhou, Lu Li, Daxiang Dong, Yi Liu, Ying Chen, Wayne Xin Zhao, Dianhai Yu, Hua Wu. Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2018.

The paper presents the control system for a light and small pan-tilt based on multirotor unmanned aerial vehicle (UAV) for aerial remote sensing applications. The design of the pan-tilt control system consists of two main components, i.e., the hardware circuit and the software program. The hardware circuit is composed of a digital circuit system and an analog circuit system. In the design, to promote the system performance, a hardware control system based on digital signal processing (DSP) and field programmable gate array (FPGA) is proposed. Then, the components and functions of each circuit system are presented, respectively. Eventually, the system is integrated and the debugging of the software and hardware are completed.

Inertially stabilized platform (ISP) is an important component of the aerial remote sensing system, which is used to make the imaging sensors stable so as to obtain the high quality images by disturbance rejection. To improve the stabilization accuracy of an ISP under multi-source disturbance environment, an active disturbance rejection control (ADRC) control scheme is prosed, which is composed of three components, i.e., a tracking differentiator (TD), a nonlinear state error feedback (NLSEF) and an extended state observer (ESO). To verify the method, the experiments are carried out. The results show that the proposed method can improve the stabilization accuracy due to significant ability in disturbances rejection.

Synthetic aperture radar imaging requires precise position information provided by position and orientation system (POS), especially sensitive to high-order errors of position information, whereas high-order errors of global positioning system (GPS) are normally ignored in POS solution procedure. For high-precision POS, high-order errors become a significant error source with decisive effects on the accuracy of POS. In this paper, we propose a GPS high-order position errors suppression and cancellation method, which includes the traditional GPS/inertia navigation system (INS) integrated method used as the high-frequency filter, and the low-frequency filter is based on least squares support vector machine. The high- and low-frequency dual filters are constructed to suppress and eliminate high-order error. In verifying our dual-rate hybrid filtering method, POS flight experiments were conducted in November and October up to December in 2015, with results showing that the high-order position accuracy of high-precision POS (0.01°/h gyro drift) is better than 0.006 m, and the proposed method has a better performance compared with other methods.

To establish an automatic conversation system between human and computer is regarded as one of the most hardcore problems in computer science. It requires interdisciplinary techniques in information retrieval, natural language processing, and data management, etc. The challenges lie in how to respond like a human, and to maintain a relevant, meaningful, and continuous conversation. The arrival of big data era reveals the feasibility to create such a system empowered by data-driven approaches. We can now organize the conversational data as a chat companion. In this paper, we introduce a chat companion system, which is a practical conversation system between human and computer as a real application. Given the human utterances as queries, our proposed system will respond with corresponding replies retrieved and highly ranked from a massive conversational data repository. Note that 'practical' here indicates effectiveness and efficiency: both issues are important for a real-time system based on a massive data repository. We have two scenarios of single-turn and multi-turn conversations. In our system, we have a base ranking without conversational context information (for single-turn) and a context-aware ranking (for multi-turn). Both rankings can be conducted either by a shallow learning or deep learning paradigm. We combine these two rankings together in optimization. In the experimental setups, we investigate the performance between effectiveness and efficiency for the proposed methods, and we also compare against a series of baselines to demonstrate the advantage of the proposed framework in terms of [email protected], MAP, and nDCG. We present a new angle to launch a practical online conversation system between human and computer.

Algorithms based on spectral graph cut objectives such as normalized cuts, ratio cuts and ratio association have become popular in recent years because they are widely applicable and simple to implement via standard eigenvector computations. Despite strong performance for a number of clustering tasks, spectral graph cut algorithms still suffer from several limitations: first, they require the number of clusters to be known in advance, but this information is often unknown a priori; second, they tend to produce clusters with uniform sizes. In some cases, the true clusters exhibit a known size distribution; in image segmentation, for instance, human- segmented images tend to yield segment sizes that follow a power-law distribution. In this paper, we propose a general framework of power-law graph cut algorithms that produce clusters whose sizes are power-law distributed, and also does not fix the number of clusters upfront. To achieve our goals, we treat the Pitman-Yor exchangeable partition probability function (EPPF) as a regularizer to graph cut objectives. Because the result- ing objectives cannot be solved by relaxing via eigenvectors, we derive a simple iterative algorithm to locally optimize the objectives. Moreover, we show that our proposed algorithm can be viewed as performing MAP inference on a particular Pitman-Yor mixture model. Our experiments on various data sets show the effectiveness of our algorithms.

Social events are events that occur between people where at least one person is aware of the other and of the event taking place. Extracting social events can play an important role in a wide range of applications, such as the construction of social network. In this paper, we introduce the task of social event extraction for tweets, an important source of fresh events. One main challenge is the lack of information in a single tweet, which is rooted in the short and noise-prone nature of tweets. We propose to collectively extract social events from multiple similar tweets using a novel factor graph, to harvest the redundance in tweets, i.e., the repeated occurrences of a social event in several tweets. We evaluate our method on a human annotated data set, and show that it outperforms all baselines, with an absolute gain of 21% in F1.