# vae Published articles for vae. This is one page of public article previews, not the complete archive. Follow Next page to continue. Summaries are not the original full articles. ## Provably Learning Diffusion Models under the Manifold Hypothesis: Collapse and Refine DevFeed: [Provably Learning Diffusion Models under the Manifold Hypothesis: Collapse and Refine]() Original publisher: [Read original article]() Author: Wei Huang, Andi Han, Mingyuan Bai, Huanjian Zhou, Qixin Zhang, Taiji Suzuki, Kenji Fukumizu Published: 2026-11-09T05:00:00Z Content type: article Language: en Sources: [arXiv: Machine Learning]() Topics: [Machine learning](), [Representation learning]() Tags: [cs-ai](), [cs-lg](), [denoising](), [diffusion](), [diffusion-models](), [dimension](), [experiments](), [framework](), [kl-regularization](), [manifold-hypothesis](), [projection](), [quality](), [regression](), [sampling](), [vae]() ### AI overview The paper analyzes a two-stage collapse-and-refine approach to learning diffusion models under the manifold hypothesis. It argues that the small-noise score's singular normal component guides denoising toward manifold projection, while a residual correction captures density and geometry. Under stated assumptions, the authors prove recovery of the population denoiser, bound geometric alignment and residual-learning errors with rates governed by intrinsic dimension, and give an end-to-end sampling guarantee under additional structural assumptions. Their practical method, Score-induced Latent Diffusion (SiLD), learns a representation through low-noise denoising and then applies latent diffusion without KL regularization. On Stacked MNIST and molecular generation benchmarks, it achieves competitive quality against architecture-matched VAE-based latent diffusion baselines. ### Source excerpt arXiv:2605.20235v2 Announce Type: replace Abstract: Diffusion models generate high-dimensional data with remarkable quality, yet how training discovers and exploits low-dimensional manifold structure remains incompletely understood. We analyze a two-stage collapse-and-refine procedure motivated by the small-noise score: its singular normal component identifies manifold projection as the leading denoising target, while the remaining correction encodes density and geometry. Under suitable assumptions, we prove that annealed mean-field training asymptotically recovers the population denoiser and quantify the resulting geometric alignment error. With the learned representation fixed, residual score learning reduces to convex random-feature regression. We establish generalization bounds that separate residual learning error, propagated projection error, and geometric approximation bias, with statistical exponents governed by the intrinsic dimension. Under additional structural assumptions, we further derive an end-to-end sampling guarantee. Motivated by this analysis, we develop Score-induced Latent Diffusion (SiLD), a practical two-stage framework combining low-noise denoising representation learning with latent diffusion, without requiring KL regularization of the latent representation. Experiments on Stacked MNIST and molecular generation benchmarks demonstrate competitive generation quality relative to architecture-matched VAE-based latent diffusion baselines. ## VTAM: Efficient Visuo-Tactile World-Action Models for Contact-Rich Manipulation without Tactile Pretraining DevFeed: [VTAM: Efficient Visuo-Tactile World-Action Models for Contact-Rich Manipulation without Tactile Pretraining]() Original publisher: [Read original article]() Author: Haoran Yuan, Weigang Yi, Zhenyu Zhang, Wendi Chen, Yuchen Mo, Jiashi Yin, Xinzhuo Li, Yue Luo, Jialin Li, Xiangyu Zeng, Chuan Wen, Cewu Lu, Katherine Driggs-Campbell, Ismini Lourentzou Published: 2026-11-06T05:00:00Z Content type: article Language: en Sources: [arXiv: Artificial Intelligence](), [arXiv: Machine Learning]() Topics: [contact-rich manipulation](), [robot sense of touch](), [World models]() Tags: [contact-rich-manipulation](), [cs-ai](), [cs-cv](), [cs-lg](), [cs-ro](), [fine-tuning](), [flow](), [inference](), [latency](), [manipulation](), [pretraining](), [robot](), [sensors](), [simulation](), [transformer](), [vae](), [vision-language-action](), [world-model](), [zero-shot]() ### AI overview VTAM is a visuo-tactile world-action model for contact-rich manipulation that uses task demonstrations without tactile or action pretraining. It first adapts a pretrained video backbone to predict future visual and tactile latents, then fine-tunes it with an action expert that predicts actions, proprioceptive states, and a flow-force proxy from tactile optical flow. The abstract reports better results than all baselines on four real-robot tasks across two platforms and higher average success than baselines without tactile pretraining on eight UniVTAC tasks. Its reported UniVTAC success is comparable to models trained on over 30,000 hours of largely visuo-tactile data: 83.4% versus 84.5% for N0-TWAM. At inference, it is about 6x faster than N0-TWAM and 16x faster than LingBot-VA without compilation; compiled latency is comparable to Fast-WAM. Ablations found that bypassing tactile world modeling failed on every tested task, while the flow-force proxy helped prevent the policy from ignoring touch. The tactile encoder transferred zero-shot between GelSight Mini and Xense Photon sensors. ### Source excerpt arXiv:2603.23481v2 Announce Type: replace-cross Abstract: Tactile vision-language-action models (VLAs) and tactile world-action models (WAMs) are a promising new paradigm for contact-rich manipulation, but existing models either underutilize touch, rely on large-scale tactile pretraining, or incur high inference latency due to iterative video generation. We present VTAM, an efficient visuo-tactile world-action model for contact-rich manipulation that requires no tactile or action pretraining. RGB and tactile observations share a frozen video VAE and the pretrained video transformer weights. VTAM first adapts the video backbone to jointly predict future visual and tactile latents, then fine-tunes it end-to-end with a randomly initialized action expert that predicts actions, proprioceptive states, and a light flow-force proxy computed from tactile optical flow. Trained only on task demonstrations, VTAM outperforms all baselines on four real-robot tasks across two platforms with different tactile sensors, outperforms all baselines without tactile pretraining in average success on eight UniVTAC simulation tasks, and is comparable to tactile VLAs and WAMs pretrained on over 30K hours of largely visuo-tactile robot data (83.4% vs. 84.5% for N0-TWAM). Without full video denoising at inference, VTAM is approximately 6x and 16x faster than N0-TWAM and LingBot-VA without compilation, with compiled latency comparable to Fast-WAM. Ablations show that feeding frozen-VAE tactile features directly to the action expert, bypassing tactile world modeling, fails on every tested task, that Stage I visuo-tactile world-model fine-tuning improves UniVTAC success on most tasks, and that flow-force proxy keeps the policy from fully ignoring touch. The tactile encoder transfers zero-shot across GelSight Mini and Xense Photon sensors. ## Exploring the Privacy--Size Tradeoff for Differentially Private Self-Attentive Variational Autoencoders DevFeed: [Exploring the Privacy--Size Tradeoff for Differentially Private Self-Attentive Variational Autoencoders]() Original publisher: [Read original article]() Author: Rohit Sadavarte, Dhruvi Kothari, Sparsh Gupta, Eric Van Oeveren Published: 2026-11-05T05:00:00Z Content type: article Language: en Sources: [arXiv: Cryptography and Security]() Topics: [Differential Privacy](), [VAE]() Tags: [attention](), [autoencoders](), [cs-ce](), [cs-cr](), [cs-lg](), [dataset](), [datasets](), [differential-privacy](), [privacy](), [size](), [synthetic](), [synthetic-data](), [tabular-data](), [utility](), [vae]() ### AI overview SAVANT is a tabular-data variational autoencoder combining self-attention with DP-SGD. The study compares it with PrivBayes and MST across privacy budgets and dataset sizes on two public datasets, measuring the downstream classifier utility lost when trained on synthetic instead of real data. Under the same differential privacy guarantee, more training data generally improves utility. The authors report that increased data size offsets much of the cost of stricter privacy for PrivBayes and MST; a noise-scaling argument suggests DP-SGD trades data for privacy budget less favorably. Two membership-inference attacks, assessed against a measured null, found no aggregate membership signal. ### Source excerpt arXiv:2611.05582v1 Announce Type: cross Abstract: Data tends to be most useful when it is shared, and the data most worth sharing is often the data that privacy obligations restrict. Synthetic data from a generator trained under differential privacy (DP) is one way to reconcile the two, but its utility depends on the privacy budget $\varepsilon$ and on how much data the generator sees - a parameter that most data sharing applications of synthetic data control. We present SAVANT, a self-attentive variational autoencoder for tabular data trained with DP-SGD that, to our knowledge, is the first to combine attention, a VAE, and DP in one tabular model. We evaluate it against two marginal mechanisms, PrivBayes and MST, jointly over the privacy budget and the dataset size on two public datasets, measuring the utility a downstream classifier loses by training on synthetic rather than real data. We find that, under the same DP guarantee, more data generally means more utility. For the marginal mechanisms it absorbs most of what a strict budget costs, and a noise-scaling argument suggests DP-SGD exchanges data fo budget at a worse rate. Two independent membership-inference attacks, judged against a measured null, found no aggregate membership signal. ## Latent Cluster Separation as a Regularizer for Lifelong Anomaly Detection DevFeed: [Latent Cluster Separation as a Regularizer for Lifelong Anomaly Detection]() Original publisher: [Read original article]() Author: Sabarish Radhakrishnan, Nathalie Japkowicz Published: 2026-11-04T05:00:00Z Content type: article Language: en Sources: [arXiv: Machine Learning]() Topics: [Machine learning](), [VAE](), [Model Development]() Tags: [anomaly-detection](), [cluster](), [computer-science](), [cs-lg](), [detection](), [machine-learning](), [training](), [vae]() ### AI overview In a VAE-based lifelong anomaly detector, latent cluster separation can improve detection quality and backward transfer when used as a training-time regularizer while keeping the existing anomaly scoring unchanged. The authors separate this effect from centroid-distance scoring, which is often introduced alongside the regularizer, and attribute much of the method's instability to switching to distance-based scoring. Whether the regularizer performs better than that scoring alternative depends on the dataset's underlying representation. ### Source excerpt arXiv:2611.03652v1 Announce Type: new Abstract: Lifelong (continual) anomaly detection systems must learn a sequence of tasks without forgetting how to recognize anomaly types seen earlier -- a property known as backward transfer. A common way to try to improve backward transfer is to reshape a model's internal representation around learnable concept centroids, typically coupled with a new way of scoring anomalies based on distance to those centroids. Because these two design choices are usually introduced together, it is unclear which one is actually responsible for any performance gain. In this paper, we unbundle the two mechanisms within a VAE-based lifelong anomaly detector and study them separately. We find that used purely as a training-time regularizer -- while leaving anomaly scoring unchanged -- latent cluster separation can improve detection quality and backward transfer, though whether it outperforms the alternative scoring approach depends on the dataset's underlying representation. Switching to distance-based scoring is instead where most of the instability commonly associated with this class of method actually originates; we identify and explain the underlying mechanism. This benefit depends on the dataset's underlying representation, and we report this as practical guidance for choosing between the two design choices, rather than as a list of caveats. ## Benchmarking Unsupervised Jamming Detection in 5G Open RAN DevFeed: [Benchmarking Unsupervised Jamming Detection in 5G Open RAN]() Original publisher: [Read original article]() Author: Sneihil Gopal, Fernando J. Cintron, Richard A. Rouil Published: 2026-11-04T05:00:00Z Content type: article Language: en Sources: [arXiv: Artificial Intelligence](), [arXiv: Machine Learning]() Topics: [5G](), [threat detection](), [Ground truth / benchmark quality]() Tags: [5g](), [attack](), [autoencoder](), [benchmarking](), [coherence](), [cs-ai](), [cs-lg](), [detection](), [environments](), [evaluation](), [fixed](), [long-short-term-memory](), [lstm](), [open](), [telemetry](), [vae](), [validation](), [variable](), [yields]() ### AI overview The benchmark finds that a Statistical Ensemble Method (SEM) detector outperformed Vanilla, LSTM, and VAE reconstruction-based baselines on KPI telemetry from ten heterogeneous 5G environments. In leave-one-out evaluation, SEM scored 0.906 on a composite measure combining detection confidence and temporal interval coherence, versus 0.688 for the best autoencoder per fold; ground-truth validation gave SEM a mean F1 of 0.918 versus 0.521 for the best AE variant. Composite rankings matched F1 rankings across environments, but the measure failed in within-trace and cross-trace evaluations when detectors flagged far fewer samples than the jamming prevalence. The authors therefore qualify its ranking reliability to cases where detection rate matches expected prevalence. ### Source excerpt arXiv:2611.04369v1 Announce Type: cross Abstract: Radio Frequency (RF) jamming threatens fifth-generation (5G) networks. Open Radio Access Network (Open RAN) telemetry enables jamming detection without dedicated spectrum monitoring equipment, but labeled attack data are rarely available across deployments, motivating unsupervised detectors. Comparing such detectors across algorithmic families, however, remains difficult without ground-truth labels. This paper benchmarks statistical and reconstruction-based unsupervised jamming detectors using a symmetric evaluation interface that combines detection confidence and temporal interval coherence into a detector-independent composite score. We compare a Statistical Ensemble Method (SEM) detector against Vanilla, Long Short-Term Memory (LSTM), and Variational Autoencoder (VAE) baselines on Key Performance Indicator (KPI) telemetry from ten heterogeneous 5G environments with fixed and variable jamming bursts and dynamic jammer gains. Under leave-one-out (LOO) evaluation, SEM achieves a mean composite score of 0.906, compared with 0.688 for the best autoencoder (AE) per fold. Ground-truth validation yields a mean F1-score of 0.918 for SEM, compared with 0.521 for the best AE variant. The composite ranks SEM first in all environments, consistent with F1 rankings. However, in both within-trace and cross-trace evaluation, it fails for detectors that flag far fewer samples than the jamming prevalence, indicating that rankings are reliable only when detection rate matches the expected prevalence. ## Latent Anchor-Driven Test Generation for Deep Neural Networks DevFeed: [Latent Anchor-Driven Test Generation for Deep Neural Networks]() Original publisher: [Read original article]() Author: Bin Duan, Matthew B. Dwyer, Guowei Yang Published: 2026-10-30T04:00:00Z Content type: article Language: en Sources: [arXiv: Machine Learning](), [arXiv: Software Engineering]() Topics: [Deep neural networks](), [Test automation](), [VAE](), [Neural Network]() Tags: [cs-lg](), [cs-se](), [deep-neural-networks](), [exploration](), [framework](), [generation](), [mutation](), [quantization](), [safety](), [security](), [semantic](), [semantic-drift](), [space](), [test](), [vae]() ### AI overview Latte is a black-box DNN testing framework that generates test inputs by encoding seed inputs with a pre-trained VQ-VAE, applying a one-step latent mutation toward anchors sampled from alternative classes, then quantizing and decoding. In evaluations across five datasets and ten models, including single- and multi-model testing, the authors report greater fault exposure and behavioral diversity than comparison methods under matched budgets. In single-model testing, Latte also maintained low semantic drift relative to the seed inputs. The supplied text does not specify the baselines or quantitative results. ### Source excerpt arXiv:2606.04310v2 Announce Type: replace-cross Abstract: Deep Neural Networks (DNNs) are increasingly being deployed in security-critical and safety-sensitive applications, which makes rigorous testing essential to identify and mitigate model weaknesses. Existing DNN testing approaches explore either the input space or a learned latent space. While latent-space generation can better maintain plausibility than direct input-space mutation, current methods still face a trade-off among exploration controllability, failure diversity, and seed-relative semantic drift. To overcome these limitations, we propose Latte, a black-box testing framework that generates semantically proximate, diverse, and fault-revealing test cases by leveraging the latent space. Specifically, Latte encodes each input seed with a pre-trained VQ-VAE and performs a seed-centered, one-step latent mutation along directions defined by anchors sampled from alternative classes, followed by quantization and decoding back to the input space. This explores local neighborhoods around each seed within the learned latent manifold, resulting in a larger number and broader diversity of oracle-triggering prediction discrepancies under the same budget. We evaluated Latte on 5 datasets and 10 DNN models in single-model and multi-model testing scenarios. Across the evaluated datasets and models, Latte improves fault exposure and behavioral diversity under matched testing budgets. Under the single-model setting, it also maintains low seed-relative semantic drift with respect to the source seeds. ## Multi-output LSTM soft sensors for industrial anaerobic co-digestion using LSTM-beta-VAE generative data augmentation DevFeed: [Multi-output LSTM soft sensors for industrial anaerobic co-digestion using LSTM-beta-VAE generative data augmentation]() Original publisher: [Read original article]() Author: Rohit Murali, Benaissa Dekhici, Tao Chen, Dongda Zhang, Michael Short Published: 2026-10-30T04:00:00Z Content type: article Language: en Sources: [arXiv: Artificial Intelligence](), [arXiv: Machine Learning]() Topics: [data augmentation](), [AI Models](), [VAE]() Tags: [abstract](), [algorithms](), [cs-ai](), [cs-lg](), [cs-sy](), [data-augmentation](), [data-driven](), [dataset](), [datasets](), [early-warning](), [eess-sy](), [failure](), [feature](), [framework](), [generative](), [hybrid](), [industrial](), [long-short-term-memory](), [lstm](), [mo](), [model](), [modelling](), [sensors](), [vae]() ### AI overview Industrial anaerobic digestion data are sparse and weighted toward steady operation, making Random Forest, XGBoost, and unaugmented LSTMs poor at recognizing transitions to instability and estimating stability indicators. The study combines an LSTM-beta-Variational Autoencoder with a multi-output LSTM soft sensor, using one year of full-scale plant data to synthesize physically valid sequences of unstable transitions. The augmented model reportedly estimated all targets simultaneously and robustly, offering a basis for early-warning soft sensors. The supplied text provides no numerical performance results. ### Source excerpt arXiv:2610.39768v1 Announce Type: cross Abstract: Anaerobic digestion (AD) requires close monitoring of key process variables, including pH, the ratio of volatile organic acids to alkaline buffer capacity and biogas production, to maintain biological stability and prevent process failure. While mechanistic frameworks such as the Anaerobic Digestion Model No. 1 are highly parameterised, computationally expensive, and difficult to calibrate for real-time applications, data-driven Long Short-Term Memory (LSTM) networks offer a potentially more adaptable alternative. However, current data-driven approaches in AD are predominantly limited to single-output architectures that focus solely on production yields, as opposed to simultaneous prediction of vital stability indicators. Furthermore, industrial AD datasets are typically sparse and skewed toward steady-state operations. Baseline evaluations demonstrated this scarcity prevents standard algorithms (Random Forest, XGBoost) and unaugmented LSTMs from recognising the transition to unstable operations, resulting in poor estimation of stability indicators. To overcome these challenges, this study proposes a novel generative data augmentation framework using an LSTM-beta-Variational Autoencoder coupled with a multi-output LSTM (MO-LSTM) soft sensor. By synthesising physically valid sequences of these transitions to unstable operations from one year of full-scale plant data, the method safely populates sparse regions of the feature space. Trained on this augmented dataset, the MO-LSTM achieved robust simultaneous estimation across all targets. This approach mitigates data sparsity, providing a basis for hybrid mechanistic-data-driven modelling and early-warning soft sensors in industrial bioprocesses. ## Uncertainty-Guided Cross-Modal Alignment for Diffusion-Based IoT Intrusion Detection DevFeed: [Uncertainty-Guided Cross-Modal Alignment for Diffusion-Based IoT Intrusion Detection]() Original publisher: [Read original article]() Author: Mahdi Arab Loodaricheh, Anita Raja, Mohammad Hossein Manshaei Published: 2026-10-30T04:00:00Z Content type: article Language: en Sources: [arXiv: Machine Learning]() Topics: [Internet of things](), [VAE](), [diffusion-transformers]() Tags: [alignment](), [attack](), [cs-lg](), [diffusion](), [experiments](), [intrusion-detection](), [iot](), [multimodal](), [reliability](), [sampling](), [uncertainty](), [vae]() ### AI overview The proposed method uses normalized VAE posterior variance as a heuristic, sample-specific reliability proxy to weight directional cross-modal alignment before diffusion training. Modality-specific VAEs encode tabular flows, DeepInsight images, and packet sequences; an asymmetric stop-gradient objective aligns their latent representations, which a class-conditional EDM then models jointly to generate attack-conditioned latent blocks. On TON-IoT, this variant had the highest mean precision, recall, density, and coverage among three diffusion variants. Training-split-only augmentation produced ROC-AUC 0.9515, PR-AUC 0.8954, balanced accuracy 0.8395, and attack-class F1 0.7940 on an untouched real test set. Compared with uniform weighting under matched alignment fine-tuning, variance-based weighting improved tabular generative fidelity and downstream detection. The variance signals are described as heuristic proxies, not calibrated uncertainties. ### Source excerpt arXiv:2610.39117v1 Announce Type: new Abstract: Diffusion models evolve noisy states toward a learned data distribution, so their sampling behavior depends on the geometry of the representation in which denoising occurs. This dependence is especially important for multimodal data: independently trained encoders can map synchronized views to incompatible latent regions, while uniform alignment can let an uncertain view distort the shared geometry. To address this gap, we introduce a novel reliability-guided alignment rule that uses normalized VAE posterior variance to control directional cross-modal influence before diffusion training. We study this issue in minority-attack generation for imbalanced Internet-of-Things intrusion detection. Tabular flow attributes, a DeepInsight image, and a packet sequence are encoded by modality-specific variational autoencoders. Per-modality normalized VAE posterior variances provide bounded, sample-specific weighting signals in an asymmetric stop-gradient alignment objective. We use these quantities as heuristic encoder-derived reliability proxies rather than calibrated probabilistic uncertainties. The full modality-specific latent means obtained after alignment fine-tuning are standardized and modeled jointly by a class-conditional EDM; a second-order sampler then generates paired attack-conditioned latent blocks for the three decoders. In controlled experiments on TON-IoT, uncertainty-aware alignment obtains the highest mean precision, recall, density, and coverage among three diffusion variants. Augmenting only the training split yields ROC-AUC 0.9515, PR-AUC 0.8954, balanced accuracy 0.8395, and attack-class F1 0.7940 on an untouched real test set. Among the two variants with matched alignment fine-tuning, posterior-variance-based weighting improves both tabular generative fidelity and downstream detection relative to uniform weighting. ## An Input Skip Can Reparameterize a VAE Without Improving the Generator DevFeed: [An Input Skip Can Reparameterize a VAE Without Improving the Generator]() Original publisher: [Read original article]() Author: Kevin Lee Published: 2026-10-28T04:00:00Z Content type: article Language: en Sources: [arXiv: Machine Learning]() Topics: [VAE]() Tags: [2](), [2018](), [abstract](), [alpha](), [also](), [analysis](), [arxiv](), [assessment](), [autoencoder](), [but](), [can](), [component](), [cs-lg](), [dart](), [decoder](), [deep](), [distribution](), [generator](), [input](), [physics-ao-ph](), [skip](), [stat-ml](), [vae]() ### AI overview For a Gaussian VAE, a constant input skip can be added by scaling the decoder output and observation noise together, preserving the observation and latent distributions and exact generative scores while multiplying reconstruction MSE by the square of the scale. Independently training models may still yield different generators because optimization depends on parameter coordinates. In the reported Gaussian experiment, a skip and variance change lowered MSE but worsened energy score and an importance-sampling estimate of negative log density. Across seven annual DART records, energy-score comparisons at equal observation variance favored the skip on three 2018 records and went both ways on four original records; PPCA had lower mean energy scores than each neural fit. Estimating observation variance from training reconstruction error also worsened energy score in one fixed-encoder-and-decoder comparison. The results show reconstruction error alone does not establish generative quality. ### Source excerpt arXiv:2610.37456v1 Announce Type: new Abstract: An input skip can reduce reconstruction error while leaving a model's distribution of new samples unchanged. For a Gaussian variational autoencoder (VAE) with constant input skip $0\leq\alpha<1$, a model without a skip can be written in skip form by multiplying its decoder output by $b=1-\alpha$ and its noise variance by $b^2$. The distribution of observations and latent variables and its exact generative scores stay the same, while reconstruction mean squared error (MSE) is multiplied by $b^2$. Separate training can produce different generators because optimization depends on the parameter coordinates. In a Gaussian experiment with known density, changing skip and observation variance together reduces MSE but worsens energy score and the estimate of negative log density from importance sampling. On seven annual Deep-ocean Assessment and Reporting of Tsunamis (DART) records, energy score differences at the same observation variance have both signs on four original records and favor the skip on three records from 2018. Probabilistic principal component analysis (PPCA) has lower mean energy scores than every neural DART fit. With the encoder and decoder kept the same, estimating observation variance from training reconstruction error can also worsen energy score. Assessing generation requires more than reconstruction error. ## On the Identifiability of Hybrid Deep Generative Models: Meta-Learning as a Solution DevFeed: [On the Identifiability of Hybrid Deep Generative Models: Meta-Learning as a Solution]() Original publisher: [Read original article]() Author: Yubo Ye, Maryam Toloubidokhti, Sumeet Vadhavkar, Xiajun Jiang, Huafeng Liu, Linwei Wang Published: 2026-10-26T04:00:00Z Content type: article Language: en Sources: [arXiv: Machine Learning]() Topics: [Physics-guided deep learning](), [Machine learning](), [Deep learning]() Tags: [benchmarks](), [bias](), [component](), [cs-lg](), [deep](), [deep-learning](), [generative](), [generative-models](), [hybrid](), [learning](), [meta](), [synthetic](), [theoretical](), [theory](), [using](), [vae]() ### AI overview Hybrid deep generative models combine physics-based expressions with neural components, but their identifiability has not been theoretically established. The paper presents a theoretical analysis and proposes meta-learning formulations to construct identifiable hybrid models. On synthetic and real-data benchmarks, the authors report evidence that existing hybrid models using unconditional priors are unidentifiable, alongside strong identifiability results for their meta-learning formulations. These claims are limited to the reported benchmarks; the supplied excerpt gives no details on the formulations or evaluation methods. ### Source excerpt arXiv:2610.32748v1 Announce Type: new Abstract: The interest in leveraging physics-based inductive bias in deep learning has resulted in recent development of hybrid deep generative models (hybrid-DGMs) that integrate known physics-based mathematical expressions in neural generative models. To identify these hybrid-DGMs requires inferring parameters of the physics-based component along with their neural component. The identifiability of these hybrid-DGMs, however, has not yet been theoretically probed or established. How does the existing theory of the un-identifiability of general DGMs apply to hybrid-DGMs? What may be an effective approach to construct a hybrid-DGM with theoretically-proven identifiability? This paper provides the first theoretical probe into the identifiability of hybrid-DGMs, and presents meta-learning as a novel solution to construct identifiable hybrid-DGMs. On synthetic and real-data benchmarks, we provide strong empirical evidence for the un-identifiability of existing hybrid-DGMs using unconditional priors, and strong identifiability results of the presented meta-formulations of hybrid-DGMs. Code is available at https://github.com/GoForIt-YYB/Meta-Hybrid-VAE. ## Sculpting the Latent Space: Disentanglement with WAE-MMD and Programmable Priors DevFeed: [Sculpting the Latent Space: Disentanglement with WAE-MMD and Programmable Priors]() Original publisher: [Read original article]() Author: Quentin Fruytier, Akshay Malhotra, Shahab Hamidi-Rad, Aditya Sant, Aryan Mokhtari, Sujay Sanghavi Published: 2026-10-22T04:00:00Z Content type: article Language: en Sources: [arXiv: Machine Learning]() Topics: [Representation learning](), [VAE]() Tags: [adversarial](), [aggregate](), [alignment](), [cs-ai](), [cs-lg](), [distribution](), [entanglement](), [machine-learning](), [quality](), [quantization](), [reduction](), [vae]() ### AI overview The paper reports that VAE and adversarial-training methods fail to reliably produce factorized aggregate posteriors, leaving latent representations entangled. It measures mutual predictability with the unsupervised Latent Predictability Score (LPS), described as an empirical lower bound on entanglement, and uses Maximum Mean Discrepancy (MMD) to shape the aggregate posterior around programmable priors. The authors report 70% less detected entanglement than leading baselines, optional semi-supervised semantic alignment, and precise distribution shaping validated with quantization-aware training, without sacrificing reconstruction quality. ### Source excerpt arXiv:2510.11953v2 Announce Type: replace Abstract: Learning disentangled representations, where semantic features are independently captured in a latent space, is dominated by Variational Auto Encoders (VAE) and Adversarial Training. In this paper, we provide direct quantitative and visual evidence that these methods consistently fail to enforce the target factorized representation on the aggregate posterior, leading to entangled representations. We quantify this failure using the Latent Predictability Score (LPS), an unsupervised metric of mutual predictability that empirically lower bounds the level of entanglement in a latent space. To address these limitations, we leverage the non-parametric Maximum Mean Discrepancy (MMD) to directly sculpt the aggregate posterior enabling programmable priors. We demonstrate (1) state-of-the-art unsupervised statistical independence, with a 70% reduction in detected entanglement compared to leading baselines; (2) the flexibility to optionally inject semi-supervised information for explicit semantic alignment; and (3) precise shaping of the aggregate posterior distribution (validated through quantization-aware training) without sacrificing reconstruction quality. ## Industrial ReMEmbR: Scaling Spatiotemporal Memory to Robot Fleets DevFeed: [Industrial ReMEmbR: Scaling Spatiotemporal Memory to Robot Fleets]() Original publisher: [Read original article]() Author: Jad Yammine, Malek Wahidi, Abrar Anwar, Yan Chang, Mihir Acharya, Anthony Rizk Published: 2026-10-21T04:00:00Z Content type: article Language: en Sources: [arXiv: Software Engineering]() Topics: [Retrieval Augmented Generation (RAG)](), [Nemotron](), [Embeddings]() Tags: [cs-ai](), [cs-ro](), [cs-se](), [demand](), [embeddings](), [gdpr](), [industrial](), [jetson](), [memory](), [nemotron](), [nvidia](), [people](), [pii](), [robot](), [robot-fleets](), [server](), [split](), [storage](), [system](), [vae](), [video](), [vision]() ### AI overview IndustrialRemembr extends NVIDIA ReMEmbR into shared spatiotemporal memory for robot fleets, generating scene descriptions only when queried to reduce server-side computation. NVIDIA Jetsons encode video clips into Cosmos-Embed1 retrieval embeddings, Cosmos3-Edge vision tokens, and Wan2.2 VAE reconstruction latents without uploading raw camera frames. The system reuses its encoding models to detect people in image patches and zero them out for PII abstraction, supporting GDPR-compliant deployments. Cosmos3-Edge is split at its vision-language boundary: vision encoding runs on robots, while language decoding runs on an on-premises server. A Nemotron agent retrieves memories by content, position, and time, then queries the decoder using cached vision tokens. The authors report that cached-token decoding matched the unsplit model's output in all 450 tested decodes. ### Source excerpt arXiv:2610.28624v1 Announce Type: cross Abstract: Warehouse robot fleets continuously observe operational conditions, but preserving these observations as reusable semantic memory incurs substantial computation, communication, and storage costs. For example, explaining recurring aisle blockages can require reviewing video from multiple robots across several shifts. We present IndustrialRemembr, a shared spatiotemporal memory system that extends NVIDIA ReMEmbR to robot fleets and reduces server-side computation by generating scene descriptions only on demand. On each robot, an NVIDIA Jetson encodes video clips into retrieval embeddings with Cosmos-Embed1, vision tokens with Cosmos3-Edge, and reconstruction latents with the Wan2.2 VAE, without uploading raw camera frames. The encoding models are re-used to detect people in image patches, zeroing them out to support PII abstraction for GDPR-compliant deployments. We split Cosmos3-Edge at its vision-language boundary, placing the vision encoder on the robot and the language decoder on an on-premises server. A Nemotron agent retrieves memories by content, position, and time and queries the decoder using cached vision tokens. This design eliminates caption generation at write time, while cached-token decoding reproduces the unsplit model's output in all 450 tested decodes. ## Remote VAEs for decoding with Inference Endpoints 🤗 DevFeed: [Remote VAEs for decoding with Inference Endpoints 🤗]() Original publisher: [Read original article]() Author: hlky; Sayak Paul Published: 2025-02-24T00:00:00Z Content type: tutorial Language: en Sources: [Hugging Face]() Topics: [VAE](), [inference-endpoints](), [diffusers](), [hugging face](), [Inference](), [Open Source](), [Deadlock]() Tags: [artificial-intelligence](), [concurrency](), [cpu](), [data](), [diffusers](), [diffusion](), [generate](), [generation](), [getting-started](), [gpu](), [hdd](), [inference](), [inference-endpoints](), [nvme](), [open-source](), [queue](), [remote](), [tensors](), [time](), [vae]() ### AI overview This article presents an experimental approach for decoding latent-space diffusion outputs with remote VAEs hosted on Inference Endpoints. It explains how remote decoding can reduce consumer GPU memory pressure, avoid the quality loss associated with tiled decoding, and improve concurrency by queueing generation requests. The article includes setup and usage examples for random tensors and pipelines involving SD v1.5, Flux, and HunyuanVideo. ### Source excerpt We're on a journey to advance and democratize artificial intelligence through open source and open science. ## Vae Soli DevFeed: [Vae Soli]() Original publisher: [Read original article]() Published: 2013-06-13T00:00:00Z Content type: article Language: en Sources: [Juri Strumpflohner]() Topics: [software-development](), [Self-organizing Team]() Tags: [account](), [allows](), [also](), [blogging](), [comment](), [development](), [don-t](), [feedback](), [focus](), [free](), [go](), [great](), [i](), [integrate](), [it](), [just](), [know](), [let](), [published](), [vae]() ### AI overview The author introduces a Medium post about teamwork in software development, focusing on people who do not integrate into a team and the conflicts that can result. They describe Medium as a blogging platform with social and collaborative writing features, including inviting others to comment and provide feedback. ### Source excerpt Lorem ipsum dolor sit amet