We introduce neural information field filter, a Bayesian state and parameter estimation method for high-dimensional nonlinear dynamical systems given large measurement datasets. Solving such a problem using traditional methods, such as Kalman and particle filters, is computationally expensive. Information field theory is a Bayesian approach that can efficiently reconstruct dynamical model state paths and calibrate model parameters from noisy measurement data. To apply the method, we parameterize the time evolution state path using the span of a finite linear basis. The existing method has to reparameterize the state path by initial states to satisfy the initial condition. Designing an expressive yet simple linear basis before knowing the true state path is crucial for inference accuracy but challenging. Moreover, reparameterizing the state path using the initial state is easy to perform for a linear basis, but is nontrivial for more complex and expressive function parameterizations, such as neural networks. The objective of this paper is to simplify and enrich the class of state path parameterizations using neural networks for the information field theory approach. To this end, we propose a generalized physics-informed conditional prior using an auxiliary initial state. We show the existing reparameterization is a special case. We parameterize the state path using a residual neural network that consists of a linear basis function and a Fourier encoding fully connected neural network residual function. The residual function aims to correct the error of the linear basis function. To sample from the intractable posterior distribution, we develop an optimization algorithm, nested stochastic variational inference, and a sampling algorithm, nested preconditioned stochastic gradient Langevin dynamics. A series of numerical and experimental examples verify and validate the proposed method.

As one of the most popular parameter-efficient fine-tuning (PEFT) methods, low-rank adaptation (LoRA) is commonly applied to fine-tune large language models (LLMs). However, updating the weights of LoRA blocks effectively and expeditiously is challenging due to the long calculation path in the original model. To address this, we propose ResLoRA, an improved framework of LoRA. By adding residual paths during training and using merging approaches to eliminate these extra paths during inference, our method can achieve better results in fewer training steps without any extra trainable parameters or inference cost compared to LoRA. The experiments on NLG, NLU, and text-to-image tasks demonstrate the effectiveness of our method. To the best of our knowledge, ResLoRA is the first work that combines the residual path with LoRA. The code of our method is available at https://github.com/microsoft/LMOps/tree/main/reslora .

Despite the rapid progress of neuromorphic computing, inadequate capacity and insufficient representation power of spiking neural networks (SNNs) severely restrict their application scope in practice. Residual learning and shortcuts have been evidenced as an important approach for training deep neural networks, but rarely did previous work assess their applicability to the characteristics of spike-based communication and spatiotemporal dynamics. In this paper, we first identify that this negligence leads to impeded information flow and the accompanying degradation problem in previous residual SNNs. To address this issue, we propose a novel SNN-oriented residual architecture termed MS-ResNet, which establishes membrane-based shortcut pathways, and further prove that the gradient norm equality can be achieved in MS-ResNet by introducing block dynamical isometry theory, which ensures the network can be well-behaved in a depth-insensitive way. Thus we are able to significantly extend the depth of directly trained SNNs, e.g., up to 482 layers on CIFAR-10 and 104 layers on ImageNet, without observing any slight degradation problem. To validate the effectiveness of MS-ResNet, experiments on both frame-based and neuromorphic datasets are conducted. MS-ResNet104 achieves a superior result of 76.02% accuracy on ImageNet, which is the highest to our best knowledge in the domain of directly trained SNNs. Great energy efficiency is also observed, with an average of only one spike per neuron needed to classify an input sample. We believe our powerful and scalable models will provide a strong support for further exploration of SNNs.