This paper improves image generation models using a multi-grid method.
problem Improving image generation models.
method A multi-grid method for learning energy-based generative ConvNet models.
result The multi-grid method outperforms traditional models.
Paper proposes CoopFlow, a two-flow generator for energy-based models.
problem Training energy-based models with Langevin flow and normalizing flow.
method CoopFlow trains an energy-based model using a normalizing flow initialization and a short-run Langevin flow revision.
result CoopFlow converges to a moment matching estimator and synthesizes realistic images.
SED integrates synthesis, execution, and debugging for neural program synthesis.
problem Challenges in synthesizing complex programs that match specifications.
method SED combines synthesis, execution, and debugging to improve neural program generation.
result SED reduces error rates and outperforms standard decoding methods.
Neural model guides PBE problem solving in programming.
problem Synthesizing programs from example inputs/outputs.
method Uses a neural model to guide miniKanren's constraint logic programming system.
result Synthesizes programs faster and generalizes to larger problems.
The paper explores a non-convergent MCMC method for EBM learning.
problem Learning energy-based models using traditional methods is challenging.
method The paper uses a non-convergent, non-mixing, and non-persistent short-run MCMC to learn EBM parameters.
result The learned short-run MCMC can generate realistic images and reconstruct/interpolate between images.
Verification helps prevent model collapse when training on synthesized data.
problem Model collapse when training on generated data.
method Theoretical characterization using Gaussian mixtures, linear classifiers, and linear verifiers to assess synthesized data quality.
result Verifiers can prevent model collapse and correlate with performance.
Two new data synthesizers generate multidimensional time series for analysis.
problem Evaluate distance functions on high-dimensional time series.
method Proposed two new data synthesizers: CBF and RAM.
result Evaluation of 1-nearest neighbor classifier using DTW on generated datasets.
The paper develops methods to create private synthetic spatial point patterns.
problem Generating private synthetic spatial point patterns.
method Developed differentially private Poisson and Cox point synthesizers.
result The synthesizers effectively maintain privacy and utility of synthetic data.
Paper introduces a novel method to generate diverse inputs for neural programming by example.
problem Synthesizing programs from input/output pairs using machine learning.
method Uses an SMT solver to generate diverse input-output pairs.
result Generated inputs improve model performance and generalization.
Generative model creates unseen class examples for zero-shot learning.
problem Training and test classes overlap, leading to biased predictions.
method Variational autoencoder with feedback-driven generator and discriminator.
result Model generates and uses unseen class examples to improve classification.
InverSynth automatically tunes synthesizer parameters from audio input.
problem Manual tuning of synthesizer parameters is time-consuming and requires expertise.
method Strided convolutional neural networks for inferring synthesizer parameters.
result InverSynth outperforms baselines in synthesizer parameter tuning.
New method augments data for segmentation tasks using generative models.
problem Limited annotated data for deep learning segmentation.
method Task-driven data augmentation with generative models modeling deformation fields and intensity.
result Significantly outperforms conventional augmentation techniques in segmentation tasks.
This research improves neural synthesizers for music sounds from speech data.
problem Applying speech synthesis techniques to musical instrument sounds.
method Comparison of three neural synthesizers in three scenarios: training, zero-shot learning, and fine-tuning.
result Neural synthesizers trained on speech data and fine-tuned on music data perform better.
Synthetic speech data improves keyword spotting models with fewer real examples.
problem Training models for recognizing spoken keywords with limited real data.
method Used a pre-trained speech embedding model to extract features for training a small keyword spotting model.
result A model trained on synthetic speech data can detect 10 keywords with the same accuracy as a model trained on over 500 real examples.
Detects AI-synthesized speech using cepstral and bispectral analysis.
problem Validating the authenticity of speech from AI-generated content.
method Integrates cepstral and bispectral analysis for distinguishing human from AI-synthesized speech.
result Higher-order statistics show less correlation for human speech compared to AI-synthesis, and cepstral analysis reveals unique power components.
Accelerator synthesizes deep learning inference from C code using FPGA hardware.
problem Efficiently implementing deep learning models on FPGAs.
method Parallelized C code using Pthreads, high-level synthesis, reduced precision, zero-weight-skipping.
result Peak performance of 138 effective GOPS on VGG-16 on Intel Arria 10 SoC FPGA.
AutoMOS uses neural nets to assess speech quality without human raters.
problem Assessing the quality of synthesized speech using human raters is time-consuming and costly.
method Developed a deep recurrent neural network that uses raw waveforms as input to predict mean opinion scores (MOS).
result AutoMOS models provide utterance-level MOS estimates only slightly inferior to human raters and can be averaged for multiple utterances.
Video sequences contain rich dynamic patterns, such as dynamic texture patterns that exhibit stationarity in the temporal domain, and action patterns that are non-stationary in either spatial or temporal domain. We show that a spatial-temporal generative ConvNet can be used to model and synthesize dynamic patterns. The…
A new method controls synthesizers with normalizing flows for easier exploration and creation.
problem Complexity and parameters in modern synthesizers make them hard to master.
method Formalizing synthesizer control as latent audio space mapping with VAE and NF, disentangling flows.
result Superior performance in parameter inference and audio reconstruction, disentangling major audio variations.
We propose a categorical data synthesizer with a quantifiable disclosure risk. Our algorithm, named Perturbed Gibbs Sampler, can handle high-dimensional categorical data that are often intractable to represent as contingency tables. The algorithm extends a multiple imputation strategy for fully synthetic data by utiliz…
Auto-decoder synthesizes graphs from latent codes.
problem Creating new graph structures from specified distributions.
method Generative model learns latent codes from empirical distribution. Self-attention identifies likely connectivity patterns. Graph-based normalizing flows sample latent codes.
result Model outperforms state of the art by 1.5x in accuracy and 2x in speed.
PATOIS synthesizes code from natural language using learned code idioms.
problem Synthesizing general-purpose source code from natural language specifications is challenging.
method PATOIS uses a neural synthesizer that interleaves high-level and low-level reasoning, incorporating learned code idioms from a corpus.
result Using learned code idioms improves the synthesizer's accuracy on semantic parsing datasets.
SYNTHONY selects tabular synthesizers based on stress profiling and user intent.
problem Non-uniform performance of tabular generative models across datasets.
method Stress profiling and intent-conditioned tabular synthesis selection.
result Meta-features predict synthesizer performance, improving selection accuracy.
This paper synthesizes mp-MRI data for prostate cancer detection.
problem Synthesizing clinically significant prostate cancer MRI data.
method Semi-supervised adversarial learning with StitchLayer and auxiliary distance maximization.
result Synthesizes high-quality mp-MRI images with distinguishable prostate cancer lesions.
AI system synthesizes chemical plant operation procedures for efficiency and stability.
problem Developing efficient and stable operation procedures for complex chemical plants.
method Integrates automated reasoning, deep reinforcement learning, and dynamic simulation with external knowledge.
result Synthesized procedure achieves faster recovery from malfunctions compared to standard PID control.
CMTS synthesizes near-miss driving scenarios for safer autonomous driving tests.
problem Lack of near-miss driving data for testing autonomous driving algorithms.
method Generative model conditioned on road maps, using Variational Bayesian methods.
result Synthesized data covers more near-miss scenarios, improving trajectory prediction and risk handling.
Synthesizes robust estimators for domain adaptation.
problem Improving prediction accuracy in target domain with limited data.
method Synthesizes a family of robust least squares estimators using convex optimization.
result Robust strategies can outperform non-robust interpolations.
Neural network synthesizes percussive sounds with adjustable timbral features.
problem Control over high-level timbral characteristics of percussive sounds.
method Feedforward convolutional neural network mapping input parameters to waveform.
result Changing input parameters produces a waveform congruent with desired characteristics.
This research synthesizes 12-lead ECG from a single-lead ECG device.
problem Limited cardiac diagnostics from single-lead ECG devices.
method Random forest machine learning model using historical 12-lead recordings.
result Synthesized 12-lead ECG with accuracies exceeding 90%.
A new approach for specifying and synthesizing subroutines for optimizing metrics.
problem Specifying and optimizing subroutines for various metrics.
method Formalizing programming by rewards (PBR), using continuous-optimization techniques to synthesize decision functions as if-then-else programs.
result Synthesized decision functions are optimal in cases when rewards have nice properties.
Paper trains two neural networks for image synthesis and learning.
problem Training two neural networks for image synthesis and learning.
method Cooperative training of a descriptor network and a generator network using maximum likelihood learning with MCMC sampling.
result The cooperative learning algorithm can train both models simultaneously and learn highly realistic generative models.
Synthesizes images from audio and visual data using spike-based autoencoders.
problem Extracting meaningful information from spatio-temporal data for image synthesis.
method Spike-based autoencoders trained to learn spatio-temporal representations of audio and visual data.
result Synthesized images from audio samples with high fidelity, achieving competitive performance.
DeepInversion generates images from trained networks without additional data.
problem Generating images from trained networks without real data.
method Inverting a trained network to synthesize images from random noise, optimizing using batch normalization layers and maximizing Jensen-Shannon divergence.
result Synthesized images from CIFAR-10 and ImageNet datasets show high fidelity and realism.
Generative adversarial network synthesizes sketches into realistic images.
problem Improving the quality and realism of facial sketches.
method Hybrid GAN with quality guided encoder and identity preserving network.
result Synthesized images are more realistic and maintain identity.
New method generates unrestricted adversarial examples from generative models.
problem Traditional defense methods are ineffective against unrestricted adversarial examples.
method Train an AC-GAN to model class-conditional distribution, then search latent space for misclassified images.
result Unrestricted adversarial examples can bypass strong defenses and fool classifiers.
Generative model synthesizes earthquake acceleration data.
problem Robust estimation of ground motions for engineering applications.
method Wasserstein GAN formulation for conditioning on physical variables.
result Trained model synthesizes realistic 3-component accelerograms.
Generative model synthesizes complex data structures with composite and nested types.
problem Synthesizing complex data structures with composite and nested types.
method Generic framework using causal transformers for struct and list generation.
result Consistently outperforms state-of-the-art models on standard and complex hierarchical datasets.
In latent Gaussian trees the pairwise correlation signs between the variables are intrinsically unrecoverable. Such information is vital since it completely determines the direction in which two variables are associated. In this work, we resort to information theoretical approaches to achieve two fundamental goals: Fir…
New method assesses graph generators using graph classifiers.
problem Quantifying how well generative models create realistic graphs.
method Using graph classifiers to evaluate synthesized graphs against real ones.
result Inability of a classifier to distinguish real from synthetic graphs indicates poor model performance.
Synthesizes faces from facial features, invariant to pose and expression.
problem Creating realistic face images from facial features.
method Learning facial landmarks and textures from facial-recognition features, training on frontal, neutral-expression images.
result Generated images are invariant to lighting, pose, and expression.
SING generates musical notes from instruments in real-time.
problem Efficiently generating high-quality audio from MIDI data.
method Frame-by-frame waveform generation with a single decoder, using a new loss function.
result SING produces significantly improved audio quality compared to state-of-the-art models, with 32x faster training and 2,500x faster inference.
Method synthesizes 4D CMR images from XCAT model using GAN and SPADE.
problem Synthesizing realistic 4D CMR images with annotations and adaptable styles.
method Hybrid GAN approach with XCAT anatomical ground truth and SPADE for semantic preservation.
result Synthesized images with modality-specific features learned from real CMR data.
PLANS synthesizes programs from noisy inputs using neural specs and filtering.
problem Synthesizing robust programs from noisy, raw inputs.
method Hybrid model combining neural extraction and rule-based synthesis with noise filtering.
result State-of-the-art performance in diverse environments with no ground-truth training.
Researchers solved a complex problem for a specific type of 4-manifolds.
problem Realizing mapping classes of finite order on del Pezzo surfaces.
method Synthesized results from reflection group theory and 4-manifold topology.
result Provided both positive and negative examples of realizability.
BUSTLE synthesizes programs by learning from intermediate values.
problem Challenges in synthesizing complex programs due to large search space.
method Bottom-up search guided by a neural network trained on input-output examples.
result Bottom-up search with execution of intermediate programs provides valuable semantic information.
PLRM synthesizes labels from mismatched sources for better training sets.
problem Creating labeled training sets is a major challenge in machine learning.
method PLRM uses probabilistic modeling to synthesize labels from indirect supervision sources with different output spaces.
result PLRM outperforms baselines by 2%-9% on various tasks.
ARF synthesizes epidemiological data to match original findings.
problem Synthetic data quality and privacy in epidemiology.
method Adversarial Random Forests (ARF) for efficient data synthesis.
result ARF-generated synthetic data consistently matches original epidemiological findings.
Method learns to synthesize eye fundus images from vessel trees.
problem Synthesizing images of the eye fundus is challenging.
method Adversarial learning technique to map vessel trees to retinal images.
result Synthetic images retain high quality of true images.