arXiv subjects

Meixi Chen

Publications and source records attributed to Meixi Chen.

6 recordsLinked to original sources

arXiv2025

Kiss3DGen: Repurposing Image Diffusion Models for 3D Asset Generation

Diffusion models have achieved great success in generating 2D images. However, the quality and generalizability of 3D content generation remain limited. State-of-the-art methods often require large-scale 3D assets for training, which are challenging to collect. In this work, we introduce Kiss3DGen (Keep It Simple and Straightforward in 3D Generation), an efficient framework for generating, editing, and enhancing 3D objects by repurposing a well-trained 2D image diffusion model for 3D generation. Specifically, we fine-tune a diffusion model to generate ''3D Bundle Image'', a tiled representation composed of multi-view images and their corresponding normal maps. The normal maps are then used to reconstruct a 3D mesh, and the multi-view images provide texture mapping, resulting in a complete 3D model. This simple method effectively transforms the 3D generation problem into a 2D image generation task, maximizing the utilization of knowledge in pretrained diffusion models. Furthermore, we demonstrate that our Kiss3DGen model is compatible with various diffusion model techniques, enabling advanced features such as 3D editing, mesh and texture enhancement, etc. Through extensive experiments, we demonstrate the effectiveness of our approach, showcasing its ability to produce high-quality 3D models efficiently.

cs.GR↗

arXiv2024

Watertox: The Art of Simplicity in Universal Attacks A Cross-Model Framework for Robust Adversarial Generation

Contemporary adversarial attack methods face significant limitations in cross-model transferability and practical applicability. We present Watertox, an elegant adversarial attack framework achieving remarkable effectiveness through architectural diversity and precision-controlled perturbations. Our two-stage Fast Gradient Sign Method combines uniform baseline perturbations ($\epsilon_1 = 0.1$) with targeted enhancements ($\epsilon_2 = 0.4$). The framework leverages an ensemble of complementary architectures, from VGG to ConvNeXt, synthesizing diverse perspectives through an innovative voting mechanism. Against state-of-the-art architectures, Watertox reduces model accuracy from 70.6% to 16.0%, with zero-shot attacks achieving up to 98.8% accuracy reduction against unseen architectures. These results establish Watertox as a significant advancement in adversarial methodologies, with promising applications in visual security systems and CAPTCHA generation.

cs.CV↗

arXiv2024

Fast and Scalable Inference for Spatial Extreme Value Models

The generalized extreme value (GEV) distribution is a popular model for analyzing and forecasting extreme weather data. To increase prediction accuracy, spatial information is often pooled via a latent Gaussian process (GP) on the GEV parameters. Inference for GEV-GP models is typically carried out using Markov chain Monte Carlo (MCMC) methods, or using approximate inference methods such as the integrated nested Laplace approximation (INLA). However, MCMC becomes prohibitively slow as the number of spatial locations increases, whereas INLA is only applicable in practice to a limited subset of GEV-GP models. In this paper, we revisit the original Laplace approximation for fitting spatial GEV models. In combination with a popular sparsity-inducing spatial covariance approximation technique, we show through simulations that our approach accurately estimates the Bayesian predictive distribution of extreme weather events, is scalable to several thousand spatial locations, and is several orders of magnitude faster than MCMC. A case study in forecasting extreme snowfall across Canada is presented.

stat.ME↗

arXiv2023

Functional Connectivity: Continuous-Time Latent Factor Models for Neural Spike Trains

Modelling the dynamics of interactions in a neuronal ensemble is an important problem in functional connectivity research. One popular framework is latent factor models (LFMs), which have achieved notable success in decoding neuronal population dynamics. However, most LFMs are specified in discrete time, where the choice of bin size significantly impacts inference results. In this work, we present what is, to the best of our knowledge, the first continuous-time multivariate spike train LFM for studying neuronal interactions and functional connectivity. We present an efficient parameter inference algorithm for our biologically justifiable model which (1) scales linearly in the number of simultaneously recorded neurons and (2) bypasses time binning and related issues. Simulation studies show that parameter estimation using the proposed model is highly accurate. Applying our LFM to experimental data from a classical conditioning study on the prefrontal cortex in rats, we found that coordinated neuronal activities are affected by (1) the onset of the cue for reward delivery, and (2) the sub-region within the frontal cortex (OFC/mPFC). These findings shed new light on our understanding of cue and outcome value encoding.

stat.ME↗

arXiv2022

Improving Few-Shot Performance of Language Models via Nearest Neighbor Calibration

Pre-trained language models (PLMs) have exhibited remarkable few-shot learning capabilities when provided a few examples in a natural language prompt as demonstrations of test instances, i.e., in-context learning. However, the performance of in-context learning is susceptible to the choice of prompt format, training examples and the ordering of the training examples. In this paper, we propose a novel nearest-neighbor calibration framework for in-context learning to ease this issue. It is inspired by a phenomenon that the in-context learning paradigm produces incorrect labels when inferring training instances, which provides a useful supervised signal to calibrate predictions. Thus, our method directly augments the predictions with a $k$-nearest-neighbor ($k$NN) classifier over a datastore of cached few-shot instance representations obtained by PLMs and their corresponding labels. Then adaptive neighbor selection and feature regularization modules are introduced to make full use of a few support instances to reduce the $k$NN retrieval noise. Experiments on various few-shot text classification tasks demonstrate that our method significantly improves in-context learning, while even achieving comparable performance with state-of-the-art tuning-based approaches in some sentiment analysis tasks.

cs.CL↗

arXiv2022

A Multivariate Point Process Model for Simultaneously Recorded Neural Spike Trains

The current state-of-the-art in neurophysiological data collection allows for simultaneous recording of tens to hundreds of neurons, for which point processes are an appropriate statistical modelling framework. However, existing point process models lack multivariate generalizations which are both flexible and computationally tractable. This paper introduces a multivariate generalization of the Skellam process with resetting (SPR), a point process tailored to model individual neural spike trains. The multivariate SPR (MSPR) is biologically justified as it mimics the process of neural integration. Its flexible dependence structure and a fast parameter estimation method make it well-suited for the analysis of simultaneously recorded spike trains from multiple neurons. The strengths and weaknesses of the MSPR are demonstrated through simulation and analysis of experimental data.

stat.ME↗