Deep learning system generates new Chinese fonts via style variables.
problem Efficiently design new Chinese fonts.
method End-to-end deep learning system generating new style fonts via interpolation of latent style-related embedding variables.
result Smooth transition between different font styles achieved.
Matching only the marginal distribution of latent style variables in factorized models fails to prevent class leakage.
problem Class leakage in factorized generative models despite matching marginal distributions.
method Derive an exact decomposition showing four conditions required for factorized sampling, and demonstrate that matching only the marginal distribution is insufficient.
result Class labels can be recovered with high accuracy (74%--100%) from factorized generative models, indicating leakage.
A new method for image translation without paired data.
problem Image-to-image translation between two domains with content preservation.
method Energy-based model in latent space of pretrained autoencoder.
result Improved translation quality and content preservation.
New method recovers diverse policies from expert data using state-action pair weighting.
problem Recovering diverse policies from expert trajectories.
method Pointwise mutual information weighted behavioral cloning.
result Effective in focusing on state-action pairs most representative of the style.
The study analyzes how data augmentation helps isolate content from style in self-supervised learning.
problem Understanding how data augmentation affects the separation of content and style in self-supervised learning.
method Formulated a latent variable model with content and style components, studied identifiability of latent representation, and introduced a dataset to test the theory.
result Sufficient conditions for identifying the invariant content partition in self-supervised learning.
Proposes a VAE variant for ordinal content factors.
problem Isolating ordinal-valued content factors in deep latent variable models.
method Introduces a partially ordered set (poset) structure and a conditional Gaussian spacing prior model.
result Significant improvements in content-style separation over previous non-ordinal approaches.
New method disentangles style features from data augmentations.
problem Difficulty in deducing which data attributes are 'style' and should be discarded.
method Structured data augmentation with multiple style embedding spaces, maximizing joint entropy.
result Empirically demonstrates benefits on synthetic and real-world data.
Style generators can invertibly generate images, useful for image enhancement and animation.
problem Generating all possible images and maintaining invertibility are challenging for GANs.
method Proposed style generators are invertible and can generate all possible images.
result Style generators outperform other GANs and Deep Image Prior as image priors.
A new framework converts EEG signals between subjects and tasks.
problem Noise and variability in EEG data hinder generalizable signal extraction.
method Contrastive Split-Latent Permutation Autoencoder (CSLP-AE) framework.
result The CSLP-AE framework enables zero-shot conversion between unseen subjects.
Generative model blends query input with latent states for structured improvisation.
problem Generating structured music from latent states of a neural network.
method Used a Variational Autoencoder (VAE) trained on a specific style corpus, and controlled blending with a noisy channel.
result Generated music with longer-term structure that blends query input with network style.
Deep learning improves handwriting style transfer and extraction.
problem Improving handwriting style transfer and extraction using deep neural networks.
method Used a deep conditioned autoencoder on IRON-OFF handwriting data-set to explore style transfer and extraction.
result Improved metrics of state-of-the-art methods by a large margin in style transfer and extraction experiments.
LostGANs generate realistic images from reconfigurable layouts and styles.
problem Learning generative models for realistic images from reconfigurable layouts and styles.
method End-to-end training of GANs with two new components: mask maps and ISLA-Norm.
result State-of-the-art performance on COCO-Stuff and Visual Genome datasets.
Transflow Learning transforms pre-trained models without retraining.
problem Transforming pre-trained models without retraining.
method Bayesian inference to warp latent vector probability distribution.
result Transforms model outputs to resemble new data without training.
Paper proposes SA-VAE for generating stylized Chinese characters.
problem Automatic generation of stylized Chinese characters is challenging.
method Proposes Style-Aware Variational Auto-Encoder (SA-VAE) to capture content and style components.
result Shows powerful one-shot/low-shot generalization ability.
Simple method disentangles content and style from pre-trained vision models.
problem Learning interpretable features in visual representations.
method Probabilistic linear entanglement model and simple disentanglement algorithm.
result Method provably disentangles content and style features.
Paper discovers and manipulates artistic styles in paintings without supervision.
problem Automatic discovery and manipulation of artistic styles in large art collections.
method Unsupervised learning using archetypal analysis on deep image representations.
result Learned dictionary of archetypal styles can be used for style interpretation and manipulation.
Method converts emotions in nonparallel speech data.
problem Lack of parallel data for speech emotion conversion.
method Unsupervised style transfer technique for nonparallel training.
result Effectiveness demonstrated on nonparallel corpora with four emotions.
Improved autoencoders guide latent sentence representations for better text generation and manipulation.
problem Current autoencoders struggle to maintain coherent latent spaces for meaningful text manipulations.
method Adversarial autoencoders with a denoising objective (DAAE) to guide latent space geometry.
result DAAE provides the best trade-off between generation quality and reconstruction capacity.
Method improves deep learning robustness to domain shifts.
problem Domain shift robustness in deep learning.
method Conditional variance regularization (CoRe) to penalize style feature changes.
result Improves predictive accuracy in domain shifts.
Unified model for audio control and style transfer.
problem Explicit control and style transfer in music generation.
method Diffusion autoencoders for semantic feature extraction, disentanglement using adversarial criterion.
result Model generates audio matching timbre targets with specified structure.
Method recombines image content and style from different images.
problem Recombining image content and style from different images.
method Constructs content embedding, uses VAE with leakage filtering to ensure separation of style and content.
result Synthesizes novel images with state-of-the-art performance on few-shot learning tasks.
Model disentangles font content and style.
problem Analyzing and reconstructing fonts.
method Variational inference and asymmetric transpose convolutional process.
result Model outperforms state-of-the-art models in font reconstruction.
New method predicts speaking style from text alone.
problem Predicting expressive speaking style from text.
method Text-Predicted Global Style Token (TP-GST) architecture.
result Synthesized speech has more pitch and energy variation.
New method reduces text classification errors by learning writing style instead of content.
problem Deep neural networks learn superficial patterns specific to training data.
method Adversarial training to unlearn confounding features.
result Model generalizes better and learns writing style features.
FIA method provides explainable recommendations for matrix factorization models.
problem Lack of explainability in latent factor models for recommendation.
method Influence functions from robust statistics to deliver neighbor-style explanations.
result FIA method successfully enforces explicit neighbor-style explanations to LFMs.
LIMP learns latent shapes with metric preservation, improving generative models.
problem Insufficient training data for high-fidelity latent representations.
method Metric preservation as a prior, geometric distortion criterion, geodesic loss.
result Synthetic samples of higher quality achieved through metric preservation.
Adversarial autoencoder improves music latent space learning.
problem Learning effective latent spaces for symbolic music data.
method Adversarial regularization with Gaussian mixtures.
result MusAE outperforms standard VAEs in reconstruction and interpolation.
Dynamic residual adapters improve performance across multiple latent domains without domain labels.
problem Overfitting to large domains and ignoring smaller ones in multi-domain learning.
method Dynamic residual adapters and augmentation strategies inspired by style transfer.
result Dynamic residual adapters significantly outperform standard models on multiple latent domains.
A new generator for GANs separates style and variation.
problem Improving GANs' ability to control and disentangle image attributes.
method Borrowing from style transfer, a new generator architecture.
result Improves GANs' quality and disentanglement of latent factors.
New method uses counterfactuals to reveal modular structure in deep generative models.
problem Challenges in manipulating deep generative models' latent representations without supervision.
method Proposes a non-statistical framework based on counterfactual manipulations.
result Modules of disentangled latent variables can be used for targeted interventions.
LADDER improves DG by reweighting domain-specific classifiers.
problem Challenges of domain generalization when causal mechanisms vary across domains.
method LADDER learns causal and style representations, reweights classifiers at inference.
result LADDER achieves gains in accuracy on various DG tasks.
MTDS improves sequence generation adaptability via latent code control.
problem Lack of adaptability in sequence generation models like RNNs.
method Hierarchical multi-task dynamical systems (MTDS) with latent code control.
result MTDS enables style transfer, interpolation, and morphing in generated sequences.
Proposes iVDFM for identifying latent factors in multivariate time series.
problem Identifying latent factors in multivariate time series with structural dynamics.
method Identifiable Variational Dynamic Factor Model (iVDFM) with iVAE-style conditioning.
result Identifiable latent factors up to permutation and component-wise affine transformations.
EncGAN learns multi-manifold structure and abstract features using an encoder.
problem Learning multi-manifold structure and abstract features in data.
method Uses an encoder to model manifold structure and invert it for generation, with a single latent space for shared abstract features.
result Successfully learns multi-manifold structure and abstract features on MNIST, 3D-chair, and UT-Zap50k datasets.
Generates outfits for e-commerce using neural networks.
problem Manual outfit creation by stylists is inefficient and not scalable.
method Multilayer neural network with visual and textual features.
result Generated outfits are preferred by users 21-34% more often.
Improves item recommendations by considering user experience evolution.
problem Current recommender systems ignore user experience evolution.
method Developed a generative HMM-LDA model to trace user evolution and interest facets.
result Significantly improved rating prediction over state-of-the-art baselines.
Proposes a VAE with a discrete bottleneck for better text generation.
problem VAEs struggle with latent variable auto-regressive decoding in text generation.
method Introduces a discretized bottleneck to enforce latent feature matching in a compact space.
result Demonstrates improved text generation capabilities across various tasks.
Method separates data into class and style factors using semi-supervised learning.
problem Separating generative factors of data into class and style vectors.
method Independent Vector Variational Autoencoders with semi-supervised learning and independence term.
result Improves classification performance and generation controllability.
The paper addresses misspecification in econometric models of discrete unobserved heterogeneity.
problem Misspecification in econometric models of discrete unobserved heterogeneity.
method Generalizing previous approaches to allow multiple latent variables, developing inference results for a k-means style estimator, and proposing information criteria for model selection.
result Over-fitting can be severe in k-means style estimators when the number of clusters is over-specified.
LADD models improve discrete diffusion for faster language generation.
problem Practical discrete diffusion models ignore cross-token dependencies, degrading performance.
method Introduces a learnable auxiliary latent channel, diffusing over the joint (token, latent) space.
result LADD models yield improvements on unconditional generation metrics.
Bayesian CycleGAN improves cycle-consistent GANs by stabilizing training and diversifying generated images.
problem Challenges in stabilizing training of cycle-consistent GANs leading to mode collapse.
method Proposes a Bayesian approach to stabilize training and diversify generated images.
result Improves per-pixel accuracy by 15% on Cityscapes semantic segmentation task and 20% on Monet2Photo style transfer.
Improved image translation using asymmetric gradient guidance.
problem Trade-off between style transformation and content preservation in diffusion models.
method Asymmetric gradient guidance to guide reverse diffusion sampling.
result Our method outperforms state-of-the-art models in image translation tasks.
AttGAN edits facial attributes by changing only what you want, preserving details.
problem Facial attribute editing with preservation of details.
method Encoder-decoder architecture with attribute classification and reconstruction learning.
result Outperforms state-of-the-arts on realistic attribute editing with preserved details.
Generative models learn latent process to match target distributions.
problem Training flow-matching models with auxiliary stochastic dynamics.
method Introduces latent process generator matching, treating generative state as a deterministic image of a Markov process.
result Learn generator of a stochastic process with same marginal distributions.
Model creates a latent space for multitrack music measures.
problem Representing and exploring the structure of polyphonic music.
method Extended MusicVAE to a latent space, enabling various operations.
result Model can generate, interpolate, and manipulate musical measures.
Bayesian model captures complex activity styles with interval relations.
problem Challenges in recognizing complex activities due to uncertainty and diversity.
method Proposes a Bayesian model using Allen's interval relations and latent variables from the Chinese restaurant process.
result Model captures all possible styles of complex activities as unique distributions over atomic actions and relations.
In this paper we address the problem of modeling relational data, which appear in many applications such as social network analysis, recommender systems and bioinformatics. Previous studies either consider latent feature based models but disregarding local structure in the network, or focus exclusively on capturing loc…
IDPGs extend RDPGs with a Poisson process for random latent positions.
problem Modeling randomness in latent positions for graph structure.
method Introduce IDPGs using Poisson point processes on latent Euclidean space.
result Continuous analogues of adjacency matrices link latent structure to observed graphs.