Search

Parameter-efficient fine-tuning (PEFT) is widely studied for its effectiveness and efficiency in the era of large language models. Low-rank adaptation (LoRA) has demonstrated commendable performance as a popular and representative method. However, it is implemented with a fixed intrinsic rank that might not be the ideal setting for the downstream tasks. Recognizing the need for more flexible downstream task adaptation, we extend the methodology of LoRA to an innovative approach we call allocating low-rank adaptation (ALoRA) that enables dynamic adjustments to the intrinsic rank during the adaptation process. First, we propose a novel method, AB-LoRA, that can effectively estimate the importance score of each LoRA rank. Second, guided by AB-LoRA, we gradually prune abundant and negatively impacting LoRA ranks and allocate the pruned LoRA budgets to important Transformer modules needing higher ranks. We have conducted experiments on various tasks, and the experimental results demonstrate that our ALoRA method can outperform the recent baselines with comparable tunable parameters., Comment: Accepted by NAACL-2024

3D Gaussian splatting, emerging as a groundbreaking approach, has drawn increasing attention for its capabilities of high-fidelity reconstruction and real-time rendering. However, it couples the appearance and geometry of the scene within the Gaussian attributes, which hinders the flexibility of editing operations, such as texture swapping. To address this issue, we propose a novel approach, namely Texture-GS, to disentangle the appearance from the geometry by representing it as a 2D texture mapped onto the 3D surface, thereby facilitating appearance editing. Technically, the disentanglement is achieved by our proposed texture mapping module, which consists of a UV mapping MLP to learn the UV coordinates for the 3D Gaussian centers, a local Taylor expansion of the MLP to efficiently approximate the UV coordinates for the ray-Gaussian intersections, and a learnable texture to capture the fine-grained appearance. Extensive experiments on the DTU dataset demonstrate that our method not only facilitates high-fidelity appearance editing but also achieves real-time rendering on consumer-level devices, e.g. a single RTX 2080 Ti GPU.

Time-dependent density functional theory (TDDFT) is widely used for understanding and predicting properties and behaviors of matter. As one of the fundamental theorems in TDDFT, van Leeuwen's theorem [Phys. Rev. Lett. 82, 3863 (1999)] guarantees how to construct a unique potential with the same one-body density evolution. Here we extend van Leeuwen's theorem by exploring truncation criteria in BBGKY-hierarchy. Our generalized theorem demonstrates the existence of a unique non-local potential to accurately reconstruct the multi-body density evolution in binary interacting systems. Under non-stringent conditions, truncation of the BBGKY-hierarchy equations aligns with the behavior of multi-body density evolution, and maintains consistency in the reduced equations. As one of applications within the extended TDDFT supported by our theorem, multiple excitation energy can be typically solved as the eigenvalue of a generalized Casida's equation. The extended TDDFT provides an accurate and first-principle framework capable of describing the kinetic processes of correlated system, including strongly coupled particle transport, multiple excitation and ionization processes.

The normal modes, i.e., the eigen solutions to the dispersion relation equation, are the most fundamental properties of a plasma, which also of key importance to many nonlinear effects such as parametric and two-plasmon decay, and Raman scattering. The real part indicates the intrinsic oscillation frequency while the imaginary part the Landau damping rate. In most of the literatures, the normal modes of quantum plasmas are obtained by means of small damping approximation (SDA), which is invalid for high-$k$ modes. In this paper, we solve the exact dispersion relations via the analytical continuation (AC) scheme, and, due to the multi-value nature of the Fermi-Dirac distribution, reformation of the complex Riemann surface is required. It is found that the change of the topological shape of the root locus in quantum plasmas is quite different from classical plasmas, in which both real and imaginary frequencies of high-$k$ modes increase with $k$ in a steeper way than the typical linear behaviour as appears in classical plasmas. As a result, the temporal evolution of a high-$k$ perturbation in quantum plasmas is dominated by the ballistic modes.

Value of cross section is a fundamental parameter to depict the transport of charged particles in matters. Due to masses of orders of magnitude higher than electrons and convenience of realistic calculation, the cross section of elastic nuclei-nuclei collision is usually treated via classical mechanics. The famous Bohr criterion was firstly proposed to judge whether the treatment via classical mechanics is reliable or not. Later, Lindhard generalized the results of Coulomb to screening potentials. Considering the increasing importance of detailed ion-ion interactions under modern simulation codes in inertial confinement fusion (ICF) researches, the validation of classical transport cross section for ion-ion interactions in a big range of parameter space is certainly required. In this work, the transport cross sections via classical mechanics under repulsive Yukawa potential are compared with those via quantum mechanics. Differences of differential cross sections are found with respect to scattering angles and velocities. Our results generally indicate that the classical picture fails at the cases of both low and high velocities, which represent a significant extension of the famous Bohr criterion and its generalized variations. Furthermore, the precise validation zones of classical picture is also analysed in this work. This work is of significant importance for benchmarking the modern ion-kinetic simulation codes in ICF researches, concerning the stopping power of $\alpha$ particles in DT fuels, ion-ion friction and viscous effects in the formation of kinetic shocks.

Knowledge of the medical decision process, which can be modeled as medical decision trees (MDTs), is critical to build clinical decision support systems. However, the current MDT construction methods rely heavily on time-consuming and laborious manual annotation. In this work, we propose a novel task, Text2MDT, to explore the automatic extraction of MDTs from medical texts such as medical guidelines and textbooks. We normalize the form of the MDT and create an annotated Text-to-MDT dataset in Chinese with the participation of medical experts. We investigate two different methods for the Text2MDT tasks: (a) an end-to-end framework which only relies on a GPT style large language models (LLM) instruction tuning to generate all the node information and tree structures. (b) The pipeline framework which decomposes the Text2MDT task to three subtasks. Experiments on our Text2MDT dataset demonstrate that: (a) the end-to-end method basd on LLMs (7B parameters or larger) show promising results, and successfully outperform the pipeline methods. (b) The chain-of-thought (COT) prompting method \cite{Wei2022ChainOT} can improve the performance of the fine-tuned LLMs on the Text2MDT test set. (c) the lightweight pipelined method based on encoder-based pretrained models can perform comparably with LLMs with model complexity two magnititudes smaller. Our Text2MDT dataset is open-sourced at \url{https://tianchi.aliyun.com/dataset/95414}, and the source codes are open-sourced at \url{https://github.com/michael-wzhu/text2dt}.

We investigate the magnetotransport of topological Dirac semimetals (DSMs) by taking into account the Lifshitz transition of the Fermi arc surface states. We demonstrate that a bulk momentum-dependent gap term, which is usually neglected in study of the bulk energy-band topology, can cause the Lifshitz transition by developing an additional Dirac cone for the surface to prevent the Fermi arcs from connecting the bulk Dirac points. As a result, the Weyl orbits can be turned off by the surface Dirac cone without destroying the bulk Dirac points. In response to the surface Lifshitz transition, the Weyl-orbit mechanism for the 3D quantum Hall effect (QHE) in topological DSMs will break down. The resulting quantized Hall plateaus can be thickness-dependent, similar to the Weyl-orbit mechanism, but their widths and quantized values become irregular. Accordingly, we propose that apart from the bulk Weyl nodes and Fermi arcs, the surface Lifshitz transition is also crucial for realizing stable Weyl orbits and 3D QHE in real materials., Comment: 7

3D single object tracking has been a crucial problem for decades with numerous applications such as autonomous driving. Despite its wide-ranging use, this task remains challenging due to the significant appearance variation caused by occlusion and size differences among tracked targets. To address these issues, we present MBPTrack, which adopts a Memory mechanism to utilize past information and formulates localization in a coarse-to-fine scheme using Box Priors given in the first frame. Specifically, past frames with targetness masks serve as an external memory, and a transformer-based module propagates tracked target cues from the memory to the current frame. To precisely localize objects of all sizes, MBPTrack first predicts the target center via Hough voting. By leveraging box priors given in the first frame, we adaptively sample reference points around the target center that roughly cover the target of different sizes. Then, we obtain dense feature maps by aggregating point features into the reference points, where localization can be performed more effectively. Extensive experiments demonstrate that MBPTrack achieves state-of-the-art performance on KITTI, nuScenes and Waymo Open Dataset, while running at 50 FPS on a single RTX3090 GPU.

3D single object tracking plays an essential role in many applications, such as autonomous driving. It remains a challenging problem due to the large appearance variation and the sparsity of points caused by occlusion and limited sensor capabilities. Therefore, contextual information across two consecutive frames is crucial for effective object tracking. However, points containing such useful information are often overlooked and cropped out in existing methods, leading to insufficient use of important contextual knowledge. To address this issue, we propose CXTrack, a novel transformer-based network for 3D object tracking, which exploits ConteXtual information to improve the tracking results. Specifically, we design a target-centric transformer network that directly takes point features from two consecutive frames and the previous bounding box as input to explore contextual information and implicitly propagate target cues. To achieve accurate localization for objects of all sizes, we propose a transformer-based localization head with a novel center embedding module to distinguish the target from distractors. Extensive experiments on three large-scale datasets, KITTI, nuScenes and Waymo Open Dataset, show that CXTrack achieves state-of-the-art tracking performance while running at 34 FPS.

Point clouds captured by depth sensors are often contaminated by noises, obstructing further analysis and applications. In this paper, we emphasize the importance of point distribution uniformity to downstream tasks. We demonstrate that point clouds produced by existing gradient-based denoisers lack uniformity despite having achieved promising quantitative results. To this end, we propose GPCD++, a gradient-based denoiser with an ultra-lightweight network named UniNet to address uniformity. Compared with previous state-of-the-art methods, our approach not only generates competitive or even better denoising results, but also significantly improves uniformity which largely benefits applications such as surface reconstruction.