arXiv subjects

Xunnong Xu

Publications and source records attributed to Xunnong Xu.

9 recordsLinked to original sources

arXiv2024

MSC: Multi-Scale Spatio-Temporal Causal Attention for Autoregressive Video Diffusion

Diffusion transformers enable flexible generative modeling for video. However, it is still technically challenging and computationally expensive to generate high-resolution videos with rich semantics and complex motion. Similar to languages, video data are also auto-regressive by nature, so it is counter-intuitive to use attention mechanism with bi-directional dependency in the model. Here we propose a Multi-Scale Causal (MSC) framework to address these problems. Specifically, we introduce multiple resolutions in the spatial dimension and high-low frequencies in the temporal dimension to realize efficient attention calculation. Furthermore, attention blocks on multiple scales are combined in a controlled way to allow causal conditioning on noisy image frames for diffusion training, based on the idea that noise destroys information at different rates on different resolutions. We theoretically show that our approach can greatly reduce the computational complexity and enhance the efficiency of training. The causal attention diffusion framework can also be used for auto-regressive long video generation, without violating the natural order of frame sequences.

cs.CV↗

arXiv2023

Bootstrap Motion Forecasting With Self-Consistent Constraints

We present a novel framework to bootstrap Motion forecasting with Self-consistent Constraints (MISC). The motion forecasting task aims at predicting future trajectories of vehicles by incorporating spatial and temporal information from the past. A key design of MISC is the proposed Dual Consistency Constraints that regularize the predicted trajectories under spatial and temporal perturbation during training. Also, to model the multi-modality in motion forecasting, we design a novel self-ensembling scheme to obtain accurate teacher targets to enforce the self-constraints with multi-modality supervision. With explicit constraints from multiple teacher targets, we observe a clear improvement in the prediction performance. Extensive experiments on the Argoverse motion forecasting benchmark and Waymo Open Motion dataset show that MISC significantly outperforms the state-of-the-art methods. As the proposed strategies are general and can be easily incorporated into other motion forecasting approaches, we also demonstrate that our proposed scheme consistently improves the prediction performance of several existing methods.

cs.CV↗

arXiv2020

Towards a Systematic Computational Framework for Modeling Multi-Agent Decision-Making at Micro Level for Smart Vehicles in a Smart World

We propose a multi-agent based computational framework for modeling decision-making and strategic interaction at micro level for smart vehicles in a smart world. The concepts of Markov game and best response dynamics are heavily leveraged. Our aim is to make the framework conceptually sound and computationally practical for a range of realistic applications, including micro path planning for autonomous vehicles. To this end, we first convert the would-be stochastic game problem into a closely related deterministic one by introducing risk premium in the utility function for each individual agent. We show how the sub-game perfect Nash equilibrium of the simplified deterministic game can be solved by an algorithm based on best response dynamics. In order to better model human driving behaviors with bounded rationality, we seek to further simplify the solution concept by replacing the Nash equilibrium condition with a heuristic and adaptive optimization with finite look-ahead anticipation. In addition, the algorithm corresponding to the new solution concept drastically improves the computational efficiency. To demonstrate how our approach can be applied to realistic traffic settings, we conduct a simulation experiment: to derive merging and yielding behaviors on a double-lane highway with an unexpected barrier. Despite assumption differences involved in the two solution concepts, the derived numerical solutions show that the endogenized driving behaviors are very similar. We also briefly comment on how the proposed framework can be further extended in a number of directions in our forthcoming work, such as behavioral calibration using real traffic video data, computational mechanism design for traffic policy optimization, and so on.

cs.MA↗

arXiv2017

Dynamically induced robust phonon transport and chiral cooling in an optomechanical system

The transport of sound and heat, in the form of phonons, can be limited by disorder-induced scattering. In electronic and optical settings the introduction of chiral transport, in which carrier propagation exhibits parity asymmetry, can remove elastic backscattering and provides robustness against disorder. However, suppression of disorder-induced scattering has never been demonstrated in non-topological phononic systems. Here we experimentally demonstrate a path for achieving robust phonon transport in the presence of material disorder, by explicitly inducing chirality through parity-selective optomechanical coupling. We show that asymmetric optical pumping of a symmetric resonator enables a dramatic chiral cooling of clockwise and counterclockwise phonons, while simultaneously suppressing the hidden action of disorder. Surprisingly, this passive mechanism is also accompanied by a chiral reduction in heat load leading to optical cooling of the mechanics without added damping, an effect that has no optical analogue. This technique can potentially improve upon the fundamental thermal limits of resonant mechanical sensors, which cannot be attained through sideband cooling.

physics.optics↗

arXiv2017

Optomechanically-induced chiral transport of phonons in one dimension

Non-reciprocal devices, with one-way transport properties, form a key component for isolating and controlling light in photonic systems. Optomechanical systems have emerged as a potential platform for optical non-reciprocity, due to ability of a pump laser to break time and parity symmetry in the system. Here we consider how the non-reciprocal behavior of light can also impact the transport of sound in optomechanical devices. We focus on the case of a quasi one dimensional optical ring resonator with many mechanical modes coupled to light via the acousto-optic effect. The addition of disorder leads to finite diffusion for phonon transport in the material, largely due to elastic backscattering between clockwise and counter-clockwise phonons. We show that a laser pump field, along with the assumption of high quality-factor, sideband-resolved optical resonances, suppresses the effects of disorder and leads to the emergence of chiral diffusion, with direction-dependent diffusion emerging in a bandwidth similar to the phase-matching bandwidth for Brillouin scattering. A simple diagrammatic theory connects the observation of reduced mechanical linewidths directly to the associated phonon diffusion properties, and helps explain recent experimental results.

quant-ph↗

arXiv2016

A quasi-mode theory of chiral phonons

The coherence properties of mechanical resonators are often limited by multiple unavoidable forms of loss -- including phonon-phonon and phonon-defect scattering -- which result in the scattering of sound into other resonant modes and into the phonon bath. Dynamic suppression of this scattering loss can lift constraints on device structure and can improve tolerance to defects in the material, even after fabrication. Inspired by recent experiments, here we introduce a model of phonon losses resulting from disorder in a whispering gallery mode resonator with acousto-optical coupling between optical and mechanical modes. We show that a typical elastic scattering mechanism of high quality factor (Q) mechanical modes flips the direction of phonon propagation via high-angle scattering, leading to damping into modes with the opposite parity. When the optical mode overlaps co-propagating high-Q and bulk mechanical modes, the addition of laser cooling via sideband-resolved damping of the mechanical mode of a chosen parity also damps and modifies the response of the bulk modes of the same parity. This, in turn, simultaneously improves the quality factor and reduces the thermal load of the counter-propagating high-Q modes, leading to the dynamical creation of a cold phononic shield. We compare our theoretical results to the recent experiments of Kim et al., and find quantitative agreement with our theory.

quant-ph↗

arXiv2016

Cooling a harmonic oscillator by optomechanical modification of its bath

Optomechanical systems show tremendous promise for high sensitivity sensing of forces and modification of mechanical properties via light. For example, similar to neutral atoms and trapped ions, laser cooling of mechanical motion by radiation pressure can take single mechanical modes to their ground state. Conventional optomechanical cooling is able to introduce additional damping channel to mechanical motion, while keeping its thermal noise at the same level, and as a consequence, the effective temperature of the mechanical mode is lowered. However, the ratio of temperature to quality factor remains roughly constant, preventing dramatic advances in quantum sensing using this approach. Here we propose an approach for simultaneously reducing the thermal load on a mechanical resonator while improving its quality factor. In essence, we use the optical interaction to dynamically modify the dominant damping mechanism, providing an optomechanically-induced effect analogous to a phononic band gap. The mechanical mode of interest is assumed to be weakly coupled to its heat bath but strongly coupled to a second mechanical mode, which is cooled by radiation pressure coupling to a red detuned cavity field. We also identify a realistic optomechanical design that has the potential to realize this novel cooling scheme.

quant-ph↗

arXiv2014

Quantum Nonlinear Optics Near Optomechanical Instabilities

Optomechanical systems provide a unique platform for observing quantum behavior of macroscopic objects. However, efforts towards realizing nonlinear behavior at the single photon level have been inhibited by the small size of the radiation pressure interaction. Here we show that it is not necessary to reach the single-photon strong-coupling regime in order to realize significant optomechanical nonlinearities. Instead, nonlinearities at the few quanta level can be achieved, even with weak-coupling, in a two-mode optomechanical system driven near instability. In this limit, we establish a new figure of merit for realizing strong nonlinearity which scales with the single-photon optomechanical coupling and the sideband resolution of the mechanical mode with respect to the cavity linewidth. We find that current devices based on optomechanical crystals, thought to be in the weak-coupling regime, can still achieve strong quantum nonlinearity; enabling deterministic interactions between single photons.

quant-ph↗

arXiv2013

Beating the standard quantum limit for force sensing with a coupled two-mode optomechanical system

Optomechanics allows the transduction of weak forces to optical fields, with many efforts approaching the standard quantum limit. We consider force-sensing using a mirror-in-the-middle setup and use two coupled cavity modes originated from normal mode splitting for separating pump and probe fields. We find that this two-mode model can be reduced to an effective single-mode model, if we drive the pump mode strongly and detect the signal from the weak probe mode. The optimal force detection sensitivity at zero frequency (DC) is calculated and we show that one can beat the standard quantum limit by driving the cavity close to instability. The best sensitivity achievable is limited by mechanical thermal noise and by optical losses. We also find that the bandwidth where optimal sensitivity is maintained is proportional to the cavity damping in the resolved sideband regime. Finally, the squeezing spectrum of the output signal is calculated, and it shows almost perfect squeezing at DC is possible by using a high quality factor and low thermal phonon-number mechanical oscillator.

quant-ph↗