arXiv subjects

Qi Qiu

Publications and source records attributed to Qi Qiu.

14 recordsLinked to original sources

arXiv2025

HAR-DoReMi: Optimizing Data Mixture for Self-Supervised Human Activity Recognition Across Heterogeneous IMU Datasets

Cross-dataset Human Activity Recognition (HAR) suffers from limited model generalization, hindering its practical deployment. To address this critical challenge, inspired by the success of DoReMi in Large Language Models (LLMs), we introduce a data mixture optimization strategy for pre-training HAR models, aiming to improve the recognition performance across heterogeneous datasets. However, directly applying DoReMi to the HAR field encounters new challenges due to the continuous, multi-channel and intrinsic heterogeneous characteristics of IMU sensor data. To overcome these limitations, we propose a novel framework HAR-DoReMi, which introduces a masked reconstruction task based on Mean Squared Error (MSE) loss. By raplacing the discrete language sequence prediction task, which relies on the Negative Log-Likelihood (NLL) loss, in the original DoReMi framework, the proposed framework is inherently more appropriate for handling the continuous and multi-channel characteristics of IMU data. In addition, HAR-DoReMi integrates the Mahony fusion algorithm into the self-supervised HAR pre-training, aiming to mitigate the heterogeneity of varying sensor orientation. This is achieved by estimating the sensor orientation within each dataset and facilitating alignment with a unified coordinate system, thereby improving the cross-dataset generalization ability of the HAR model. Experimental evaluation on multiple cross-dataset HAR transfer tasks demonstrates that HAR-DoReMi improves the accuracy by an average of 6.51%, compared to the current state-of-the-art method with only approximately 30% to 50% of the data usage. These results confirm the effectiveness of HAR-DoReMi in improving the generalization and data efficiency of pre-training HAR models, underscoring its significant potential to facilitate the practical deployment of HAR technology.

cs.LG↗

arXiv2024

Reliable Projection Based Unsupervised Learning for Semi-Definite QCQP with Application of Beamforming Optimization

In this paper, we investigate a special class of quadratic-constrained quadratic programming (QCQP) with semi-definite constraints. Traditionally, since such a problem is non-convex and N-hard, the neural network (NN) is regarded as a promising method to obtain a high-performing solution. However, due to the inherent prediction error, it is challenging to ensure all solution output by the NN is feasible. Although some existing methods propose some naive methods, they only focus on reducing the constraint violation probability, where not all solutions are feasibly guaranteed. To deal with the above challenge, in this paper a computing efficient and reliable projection is proposed, where all solution output by the NN are ensured to be feasible. Moreover, unsupervised learning is used, so the NN can be trained effectively and efficiently without labels. Theoretically, the solution of the NN after projection is proven to be feasible, and we also prove the projection method can enhance the convergence performance and speed of the NN. To evaluate our proposed method, the quality of service (QoS)-contained beamforming scenario is studied, where the simulation results show the proposed method can achieve high-performance which is competitive with the lower bound.

cs.LG↗

arXiv2024

Trapezoidal Gradient Descent for Effective Reinforcement Learning in Spiking Networks

With the rapid development of artificial intelligence technology, the field of reinforcement learning has continuously achieved breakthroughs in both theory and practice. However, traditional reinforcement learning algorithms often entail high energy consumption during interactions with the environment. Spiking Neural Network (SNN), with their low energy consumption characteristics and performance comparable to deep neural networks, have garnered widespread attention. To reduce the energy consumption of practical applications of reinforcement learning, researchers have successively proposed the Pop-SAN and MDC-SAN algorithms. Nonetheless, these algorithms use rectangular functions to approximate the spike network during the training process, resulting in low sensitivity, thus indicating room for improvement in the training effectiveness of SNN. Based on this, we propose a trapezoidal approximation gradient method to replace the spike network, which not only preserves the original stable learning state but also enhances the model's adaptability and response sensitivity under various signal dynamics. Simulation results show that the improved algorithm, using the trapezoidal approximation gradient to replace the spike network, achieves better convergence speed and performance compared to the original algorithm and demonstrates good training stability.

cs.AI↗

arXiv2024

ReLU-KAN: New Kolmogorov-Arnold Networks that Only Need Matrix Addition, Dot Multiplication, and ReLU

Limited by the complexity of basis function (B-spline) calculations, Kolmogorov-Arnold Networks (KAN) suffer from restricted parallel computing capability on GPUs. This paper proposes a novel ReLU-KAN implementation that inherits the core idea of KAN. By adopting ReLU (Rectified Linear Unit) and point-wise multiplication, we simplify the design of KAN's basis function and optimize the computation process for efficient CUDA computing. The proposed ReLU-KAN architecture can be readily implemented on existing deep learning frameworks (e.g., PyTorch) for both inference and training. Experimental results demonstrate that ReLU-KAN achieves a 20x speedup compared to traditional KAN with 4-layer networks. Furthermore, ReLU-KAN exhibits a more stable training process with superior fitting ability while preserving the "catastrophic forgetting avoidance" property of KAN. You can get the code in https://github.com/quiqi/relu_kan

cs.LG↗

arXiv2024

FLOW: Fusing and Shuffling Global and Local Views for Cross-User Human Activity Recognition with IMUs

Inertial Measurement Unit (IMU) sensors are widely employed for Human Activity Recognition (HAR) due to their portability, energy efficiency, and growing research interest. However, a significant challenge for IMU-HAR models is achieving robust generalization performance across diverse users. This limitation stems from substantial variations in data distribution among individual users. One primary reason for this distribution disparity lies in the representation of IMU sensor data in the local coordinate system, which is susceptible to subtle user variations during IMU wearing. To address this issue, we propose a novel approach that extracts a global view representation based on the characteristics of IMU data, effectively alleviating the data distribution discrepancies induced by wearing styles. To validate the efficacy of the global view representation, we fed both global and local view data into model for experiments. The results demonstrate that global view data significantly outperforms local view data in cross-user experiments. Furthermore, we propose a Multi-view Supervised Network (MVFNet) based on Shuffling to effectively fuse local view and global view data. It supervises the feature extraction of each view through view division and view shuffling, so as to avoid the model ignoring important features as much as possible. Extensive experiments conducted on OPPORTUNITY and PAMAP2 datasets demonstrate that the proposed algorithm outperforms the current state-of-the-art methods in cross-user HAR.

cs.CV↗

arXiv2016

New investigations on the transverse spin of structured optical fields

Guided waves and surface waves can be taken as two typical examples of structured optical fields with the transverse spin. Analytical derivations are developed to demonstrate that (i) guided waves also carry the transverse spin that depends on the mean direction of propagation, which may have important applications in spin-dependent unidirectional optical interfaces; (ii) the quantization form of the transverse spin is for the first time revealed, which is not obvious and related to an ellipticity; (iii) from a unified point of view, the transverse spin can be attributed to the presence of an effective rest mass of structured optical fields; (iv) the transverse spin can also be described by the spin matrix of the photon field; (v) unlike a free optical field whose spin projection on the propagation direction is the only observable, owing to the effective rest mass, the spin projection of structured optical fields on other directions is also an observable, such that one can develop an optical analogy of spintronics. A preliminary idea about the potential applications of the transverse spin is presented, but an in-depth and complete study will be presented in our next work.

quant-ph↗

arXiv2016

An extensive survey of dayside diffuse aurora based on optical observations at Yellow River Station

By using 7 years optical auroral observations obtained at Yellow River Station (magnetic latitude $76.24\,^{\circ}{\rm C}$N) at Ny-Alesund, Svalbard, we performed the first extensive survey for the dayside diffuse auroras (DDAs) and acquired observational results as follows. (1) The DDAs can be classified into two broad categories, i.e., unstructured and structured DDAs. The unstructured DDAs are mainly distributed in the morning and afternoon, but the structured DDAs predominantly occurred around the magnetic local noon (MLN). (2) The unstructured DDAs observed in morning and afternoon present obviously different properties. The afternoon ones are much stable and seldom show pulsating property. (3) The DDAs are more easily observed under geomagnetically quiet times. (4) The structured DDAsmainly show patchy, stripy, and irregular forms and are often pulsating and drifting. The drifting directions are mostly westward (with speed $\sim$5km/s), but there are cases showing eastward or poleward drifting. (5) The stripy DDAs are exclusively observed near theMLN and,most importantly, their alignments are confirmed to be consistent with the direction of ionospheric convection near the MLN. (6) A new auroral form, called throat aurora, is found to be developed from the stripy DDAs. Based on the observational results and previous studies, we proposed our explanations to the DDAs. We suggest that the unstructured DDAs observed in the morning are extensions of the nightside diffuse aurora to the dayside, but that observed in the afternoon are predominantly caused by proton precipitations. $\textit{(Abstract continues in PDF).}$

physics.space-ph↗

arXiv2016

The effect of gravitational spin-orbit coupling on the circular photon orbit in the Schwarzschild geometry

The (1, 0)+(0, 1) representation of the group SL(2, C) provides a six-component spinor equivalent to the electromagnetic field tensor. By means of the (1, 0)+(0, 1) description, one can treat the photon field in curved spacetime via spin connection and the tetrad formalism, which is of great advantage to study the gravitational spin-orbit coupling of photons. Once the gravitational spin-orbit coupling is taken into account, the traditional radius of the circular photon orbit in the Schwarzschild geometry should be replaced with two different radiuses corresponding to the photons with the helicities of +1 and -1, respectively. Owing to the splitting of energy levels induced by the spin-orbit coupling, photons (from Hawking radiations, say) escaping from a Schwarzschild black hole are partially polarized, provided that their initial velocities possess nonzero tangential components.

gr-qc↗

arXiv2012

Time operator in QFT with Virasoro constraints

Time operator is studied on the basis of field quantization, where the difficulty stemming from Pauli's theorem is circumvented by borrowing ideas from the covariant quantization of the bosonic string, i.e., one can remove the negative energy states by imposing Virasoro constraints. Applying the index theorem, one can show that in a different subspace of a Fock space, there is a different self-adjoint time operator. However, the self-adjoint time operator in the maximal subspace of the Fock space can also represent the self-adjoint time operator in the other subspaces, such that it can be taken as the single, universal time operator. Furthermore, a new insight on Pauli's theorem is presented.

quant-ph↗

arXiv2012

Comment on "Uncertainty Relation for Photons"

In a recent interesting Letter [Phys. Rev. Lett. 108, 140401 (2012)] I. Bialynicki-Birula and his coauthor have derived the uncertainty relation for the photons in three dimensions. However, some of their arguments are problematical, and this impacts their conclusion.

quant-ph↗

arXiv2011

Photonic zitterbewegung and its interpretation

In term of the volume-integrated Poynting vector, we present a quantum field-theory investigation on the zitterbewegung (ZB) of photons, and show that this ZB occurs only in the presence of virtual longitudinal and scalar photons. To present a heuristic explanation for such ZB, by assuming that the space time is sufficiently close to the flat Minkowski space, we show that the gravitational interaction can result in the ZB of photons.

quant-ph↗

arXiv2011

Zitterbewegung and its significance for the Hawking radiation

The Hawking radiation can be viewed from very different perspectives, not all of which can be proved to be rigorously equivalent to one another. On the other hand, an old interest in the zitterbewegung (ZB) of the Dirac electron has recently been rekindled by the investigations on spintronics and graphene, etc. In this letter, we show that, if particles emitted by black holes are electrons or positrons, one can also regard the Hawking radiation as a ZB process.

quant-ph↗

arXiv2011

Average and instantaneous velocities of energy of evanescent modes

Many theoretical and experimental investigations have presented a conclusion that evanescent electromagnetic modes can superluminally propagate. However, in this paper, we show that the average energy velocity of evanescent modes inside a cut-off waveguide is always less than or equal to the velocity of light in vacuum, while the instantaneous energy velocity can be superluminal, which does not violate causality according to quantum field theory: the fact that a particle can propagate over a space-like interval does preserve causality provided that here a measurement performed at one point cannot affect another measurement at a point separated from the first with a space-like interval.

quant-ph↗

arXiv2009

Photon wave function and zitterbewegung

In terms of a photon wave function corresponding to the (1, 0)+(0, 1) representation of the Lorentz group, the radiation and Coulomb fields within a source-free region can be described unitedly by a Lorentz-covariant Dirac-like equation. In our formalism, the relation between the positive- and negative-energy solutions of the Dirac-like equation corresponds to the duality between the electric and magnetic fields, rather than to the usual particle-antiparticle symmetry. The zitterbewegung (ZB) of photons is studied via the momentum vector of the electromagnetic field, which shows that only in the presence of virtual longitudinal and scalar photons, the ZB motion of photons can occur, and its vector property is described by the polarization vectors of the electromagnetic field.

quant-ph↗