autoencoder

The paper proposes LLMAE, a method to repurpose pre-trained decoder-only LLMs as continuous text autoencoders using a latent bottleneck, achieving high-fidelity reconstruction and enabling downstream tasks like image captioning.

The paper introduces GAE, a geometry-native autoencoder that creates a compact latent space for generating 3D-consistent scenes, enhancing visual quality and coherence over existing methods.

This paper investigates the link between autoencoder parameters and data statistics, proposing that parameters can function as a vector representation of data, supported by theoretical analysis and experiments on CIFAR-10 and FashionMNIST.

This paper introduces Drift Variation autoencoder, which uses conditional posterior flow matching to unify generative and representation learning, achieving high performance in controlled multimodal benchmarks.

This study uses meta-clustering on milk mid-infrared spectra to identify dairy cow groups associated with negative energy balance in early lactation, revealing five distinct clusters with varying severity.

Prof. Tom Yeh shares an interactive diagram for learning about autoencoders, part of his 'AI by Hand' series focused on multi-layer perceptrons.

This paper revisits energy-based models for tabular anomaly detection, demonstrating that combining Deep Boltzmann Machine energy scores with autoencoder reconstruction scores significantly improves performance on benchmark datasets.

ReconSpan introduces an adaptive latent tokenization method that divides text into reconstructible chunks using a backward decoder, enabling variable-length latent tokens and post-training control of granularity.

This paper introduces a physics-aware autoencoder-based latent-space framework for reduced-order forward modeling and variational parameter estimation in parametric dynamical systems, demonstrated on computational fluid dynamics benchmarks. The method enables differentiable surrogate-based inverse modeling and shows improved calibration robustness under realistic noisy or partial observations.

V-RAE proposes a video representation autoencoder that builds semantically organized latents from frozen vision representations to enhance video generation quality, convergence speed, and predictive modeling.

TailBooster is a dual-layer generative framework that synthesizes operationally valid extreme air-transport events using statistical tail extraction and autoencoder-based cleaning, significantly improving extreme-event prediction accuracy.

Kijai shares a quickly trained 2D tiny VAE for MiniMax-H3, intended to improve latent previews in ComfyUI. A better alternative trained by madebyollin is also linked.

Presents NEXUS, a lightweight foundation model with ~3M parameters pre-trained on LHC collision data, demonstrating improved downstream performance and cross-domain transfer to gravitational waves, flood forecasting, and neural activity.

ConvGRUAutoencoder combines convolutional layers with gated recurrent units to compress and reconstruct video sequences, enabling unsupervised learning from video without human labeling.

This paper proposes Hyper-Spherical Quantization (HSQ) to address codebook collapse in discretizing visual representations, achieving high-fidelity reconstruction and scalable codebook budgets up to 131,072 with 100% utilization.

CLOE is a new semi-supervised anomaly detection method combining an autoencoder with a Christoffel Function-based detector, using a novel loss function to improve representation learning. It achieves state-of-the-art results on high-dimensional tabular data while maintaining simplicity.

Proposes physical self-supervised learning, an autoencoder paradigm for label-free IMU sensing that replaces the neural decoder with a physics-based decoder, achieving up to 5x error reduction in tracking and motion capture tasks without manual labels.

FLARE is a forced latent autoencoder that discovers compact response coordinates and sparse input-dependent latent dynamics from high-dimensional observations of forced physical systems, enabling long-horizon forecasting under unseen inputs.

VisCo is a training-efficient self-compression framework that reuses a pretrained vision-language model as an intrinsic encoder for visual token compression, achieving superior performance across all compression ratios without external modules.

This paper introduces ODIN, a novel autoencoder architecture that enforces orthogonality and importance ordering of latent dimensions, recovering PCA-like interpretability in a fully non-linear regime. The method integrates geometric constraints into the training objective, theoretically grounded and empirically validated on synthetic and real-world datasets.