GB-FESO backpropagates a KL-divergence loss through a frozen conditional diffusion model's sampling trajectory to optimize system parameters so the generated ensemble matches a target free-energy surface.
super hub Tool reference
Categorical Reparameterization with Gumbel-Softmax
Tool reference. 75% of classified Pith citations use this work as a method, library, or software dependency, not as a substantive claim.
113
Pith papers citing it
3,241
external citations · Pith
Method reference
75% of classified citations
abstract
Categorical variables are a natural choice for representing discrete structure in the world. However, stochastic neural networks rarely use categorical latent variables due to the inability to backpropagate through samples. In this work, we present an efficient gradient estimator that replaces the non-differentiable sample from a categorical distribution with a differentiable sample from a novel Gumbel-Softmax distribution. This distribution has the essential property that it can be smoothly annealed into a categorical distribution. We show that our Gumbel-Softmax estimator outperforms state-of-the-art gradient estimators on structured output prediction and unsupervised generative modeling tasks with categorical latent variables, and enables large speedups on semi-supervised classification.
hub tools
citation-role summary
method 6
background 2
citation-polarity summary
claims ledger
- abstract Categorical variables are a natural choice for representing discrete structure in the world. However, stochastic neural networks rarely use categorical latent variables due to the inability to backpropagate through samples. In this work, we present an efficient gradient estimator that replaces the non-differentiable sample from a categorical distribution with a differentiable sample from a novel Gumbel-Softmax distribution. This distribution has the essential property that it can be smoothly annealed into a categorical distribution. We show that our Gumbel-Softmax estimator outperforms state-o
authors
co-cited works
representative citing papers
HeRo recursively partitions the LLM vocabulary into a hierarchy, embedding multi-bit payloads across layers so that verifiers with different keys recover only their authorized portion while preserving the original sampling distribution.
QuADA-GS learns to predict local complexity-driven Gaussian densification from low-resolution inputs and uses Hierarchical Pointer Convolution for efficient arbitrary-scale super-resolution.
Task-specific LoRA adapters in continual learning exhibit significant low-rank subspace overlap, enabling LiteLoRA's learned gating to reduce active adapters by 20-70% while matching or exceeding prior performance.
FleetAgent pairs a vector-to-embedding interface (VecFormer) with an MLLM to turn compact V2N messages into structured natural-language teleoperation assistance, cutting uplink payload 625x and improving Lingo-Judge score 16.8% on a new nuScenes-derived dataset.
EpiFormer improves epitope prediction F1 score by over 40% via early-fusion cross-attention in GNN layers and sparsity-aware objectives, while recovering known biology as emergent behavior.
Ego-METAS is a new benchmark providing unified egocentric video data, splits, features and baselines for online multimodal temporal action segmentation under hardware-representative energy constraints.
CDM amortizes SMC inference for reward-tilted discrete diffusion by training a parameterized twist function on contrastive samples with closed-form kernels.
GraphFlow uses a unified wGraph to dynamically instantiate workflows and manage KV caches for LLM agents, reporting 4.95 pp average gains and 4x memory reduction on five benchmarks.
TrustFlip weaponizes consistency-based trust defenses in vehicular collaborative perception by using physical adversarial objects to induce inconsistencies that are misattributed to benign vehicles, leading to their exclusion and reduced system performance.
MDIC uses a text-conditioned diffusion decoder and a supervised feature-mask generator on visual side information to achieve SOTA perceptual quality in distributed image compression at extremely low bitrates.
FLDD learns non-Markovian marginal and posterior distributions for the forward process so a factorized reverse process can match the target better and produce higher-quality samples in fewer steps.
DCDM replaces positional blocks with learnable semantic chunks via differentiable Chunking Attention, yielding consistent gains over block and unstructured diffusion baselines up to 1.5B parameters.
HPO enables unbiased policy optimization in hybrid action spaces by mixing differentiable simulation gradients with score-function estimates, outperforming PPO as continuous dimensions increase.
Test-time sparsity with a parallel pipeline and omnidirectional feature reuse accelerates action diffusion by 5x to 47.5 Hz while cutting FLOPs 92% with no performance loss.
A marginal estimator for blind deconvolution in solar imaging that integrates out object uncertainty to improve regularization and allow plug-and-play hyperparameter optimization.
SoftBlobGIN combines ESM-2 representations with protein contact graphs via a lightweight GNN and differentiable substructure pooling to achieve 92.8% accuracy on enzyme classification, raise binding-site AUROC to 0.983, and generate auditable structural explanations without retraining the language模型
DTSemNet gives an exact, invertible neural-network encoding of hard oblique decision trees that supports direct gradient training for both classification and regression without probabilistic softening or quantized estimators.
PhySPRING uses differentiable GNNs to learn hierarchical coarsened spring-mass topologies and parameters from observations, delivering up to 2.3x speedup on PhysTwin benchmarks and comparable robot policy success rates in zero-shot Real2Sim substitution.
AS-LoRA adaptively chooses which LoRA factor to update per layer and round using a curvature-aware second-order score, eliminating reconstruction error floors and improving performance in DP federated learning.
NEO is a probabilistic neural model that induces compositional programs as a learned Language of Thought from non-textual observations and executes them via a shared transition model to enable explanation-driven generalization.
ARCH is a hierarchical flow-based generative model that enables tractable conditional intensity computation and arbitrary conditioning for spatiotemporal event distributions.
Four continuous relaxations turn non-differentiable coverage and revisit calculations into a fully differentiable pipeline that optimizes satellite orbits via gradients and outperforms metaheuristics.
SecureRouter accelerates secure transformer inference by 1.95x via an encrypted router that selects input-adaptive models from an MPC-optimized pool with negligible accuracy loss.
citing papers explorer
-
Gradient-Based Inverse Design of Free-Energy Landscapes with Diffusion Models
GB-FESO backpropagates a KL-divergence loss through a frozen conditional diffusion model's sampling trajectory to optimize system parameters so the generated ensemble matches a target free-energy surface.
-
Selective Disclosure Watermarking for Large Language Models
HeRo recursively partitions the LLM vocabulary into a hierarchy, embedding multi-bit payloads across layers so that verifiers with different keys recover only their authorized portion while preserving the original sampling distribution.
-
Learning to Adaptively Allocate Gaussians for Arbitrary-Scale Image Super-Resolution
QuADA-GS learns to predict local complexity-driven Gaussian densification from low-resolution inputs and uses Hierarchical Pointer Convolution for efficient arbitrary-scale super-resolution.
-
When One Adapter Speaks for Many: Discovering Low-Rank Redundancy in Continual Fine-Tuning
Task-specific LoRA adapters in continual learning exhibit significant low-rank subspace overlap, enabling LiteLoRA's learned gating to reduce active adapters by 20-70% while matching or exceeding prior performance.
-
FleetAgent: Teleoperation Assistant for Autonomous Fleets via Vectorized V2N Messages
FleetAgent pairs a vector-to-embedding interface (VecFormer) with an MLLM to turn compact V2N messages into structured natural-language teleoperation assistance, cutting uplink payload 625x and improving Lingo-Judge score 16.8% on a new nuScenes-derived dataset.
-
EpiFormer: Learning Antigen-Antibody Interactions for Epitope Prediction via Geometric Deep Learning
EpiFormer improves epitope prediction F1 score by over 40% via early-fusion cross-attention in GNN layers and sparsity-aware objectives, while recovering known biology as emergent behavior.
-
Ego-METAS: Egocentric online Multimodal Energy-efficient Temporal Action Segmentation benchmark
Ego-METAS is a new benchmark providing unified egocentric video data, splits, features and baselines for online multimodal temporal action segmentation under hardware-representative energy constraints.
-
Contrastive Distribution Matching for Amortized Sequential Monte Carlo in Discrete Diffusion
CDM amortizes SMC inference for reward-tilted discrete diffusion by training a parameterized twist function on contrastive samples with closed-form kernels.
-
GraphFlow: A Graph-Based Workflow Management for Efficient LLM-Agent Serving
GraphFlow uses a unified wGraph to dynamically instantiate workflows and manage KV caches for LLM agents, reporting 4.95 pp average gains and 4x memory reduction on five benchmarks.
-
Adversarial Trust Poisoning in Vehicular Collaborative Perception
TrustFlip weaponizes consistency-based trust defenses in vehicular collaborative perception by using physical adversarial objects to induce inconsistencies that are misattributed to benign vehicles, leading to their exclusion and reduced system performance.
-
Distributed Image Compression with Multimodal Side Information at Extremely Low Bitrates
MDIC uses a text-conditioned diffusion decoder and a supervised feature-mask generator on visual side information to achieve SOTA perceptual quality in distributed image compression at extremely low bitrates.
-
Forward-Learned Discrete Diffusion: Learning how to noise to denoise faster
FLDD learns non-Markovian marginal and posterior distributions for the forward process so a factorized reverse process can match the target better and produce higher-quality samples in fewer steps.
-
Dynamic Chunking for Diffusion Language Models
DCDM replaces positional blocks with learnable semantic chunks via differentiable Chunking Attention, yielding consistent gains over block and unstructured diffusion baselines up to 1.5B parameters.
-
Policy Optimization in Hybrid Discrete-Continuous Action Spaces via Mixed Gradients
HPO enables unbiased policy optimization in hybrid action spaces by mixing differentiable simulation gradients with score-function estimates, outperforming PPO as continuous dimensions increase.
-
Test-time Sparsity for Extreme Fast Action Diffusion
Test-time sparsity with a parallel pipeline and omnidirectional feature reuse accelerates action diffusion by 5x to 47.5 Hz while cutting FLOPs 92% with no performance loss.
-
Marginal multi-object multi-frame blind deconvolution
A marginal estimator for blind deconvolution in solar imaging that integrates out object uncertainty to improve regularization and allow plug-and-play hyperparameter optimization.
-
Structural Interpretations of Protein Language Model Representations via Differentiable Graph Partitioning
SoftBlobGIN combines ESM-2 representations with protein contact graphs via a lightweight GNN and differentiable substructure pooling to achieve 92.8% accuracy on enzyme classification, raise binding-site AUROC to 0.983, and generate auditable structural explanations without retraining the language模型
-
Approximation-Free Differentiable Oblique Decision Trees
DTSemNet gives an exact, invertible neural-network encoding of hard oblique decision trees that supports direct gradient training for both classification and regression without probabilistic softening or quantized estimators.
-
PhySPRING: Structure-Preserving Reduction of Physics-Informed Twins via GNN
PhySPRING uses differentiable GNNs to learn hierarchical coarsened spring-mass topologies and parameters from observations, delivering up to 2.3x speedup on PhysTwin benchmarks and comparable robot policy success rates in zero-shot Real2Sim substitution.
-
Adaptive Selection of LoRA Components in Privacy-Preserving Federated Learning
AS-LoRA adaptively chooses which LoRA factor to update per layer and round using a curvature-aware second-order score, eliminating reconstruction error floors and improving performance in DP federated learning.
-
Learning to Theorize the World from Observation
NEO is a probabilistic neural model that induces compositional programs as a learned Language of Thought from non-textual observations and executes them via a shared transition model to enable explanation-driven generalization.
-
Arbitrarily Conditioned Hierarchical Flows for Spatiotemporal Events
ARCH is a hierarchical flow-based generative model that enables tractable conditional intensity computation and arbitrary conditioning for spatiotemporal event distributions.
-
Differentiable Satellite Constellation Configuration via Relaxed Coverage and Revisit Objectives
Four continuous relaxations turn non-differentiable coverage and revisit calculations into a fully differentiable pipeline that optimizes satellite orbits via gradients and outperforms metaheuristics.
-
SecureRouter: Encrypted Routing for Efficient Secure Inference
SecureRouter accelerates secure transformer inference by 1.95x via an encrypted router that selects input-adaptive models from an MPC-optimized pool with negligible accuracy loss.
-
VisPCO: Visual Token Pruning Configuration Optimization via Budget-Aware Pareto-Frontier Learning for Vision-Language Models
VisPCO uses continuous relaxation, straight-through estimators, and budget-aware Pareto-frontier learning to automatically discover optimal visual token pruning configurations that approximate grid-search results across VLMs and benchmarks.
-
In-Context Symbolic Regression for Robustness-Improved Kolmogorov-Arnold Networks
In-context symbolic regression methods improve robustness of symbolic formula recovery from KANs, cutting median OFAT test MSE by up to 99.8 percent across hyperparameter sweeps.
-
Differentiable Semantic ID for Generative Recommendation
DIGER makes semantic IDs in generative recommendation differentiable via Gumbel noise and decay schedules, yielding consistent gains on public datasets by aligning indexing and recommendation losses.
-
DEFT: Differentiable Automatic Test Pattern Generation
DEFT turns ATPG into differentiable optimization with a reparameterization and CUDA kernel, cutting test pattern counts by 27.3% on average versus commercial tools.
-
End2Reg: Learning Task-Specific Segmentation for Markerless Registration in Spine Surgery
End2Reg jointly optimizes task-specific segmentation and markerless registration via deep learning guided solely by the registration objective, cutting median target registration error by 32% and RMSE by 61% on ex- and in-vivo spine surgery benchmarks.
-
TokenChain: A Discrete Speech Chain via Semantic Token Modeling
TokenChain demonstrates that a discrete semantic-token interface can sustain effective chain learning between ASR and TTS, yielding faster convergence and lower error rates on LibriSpeech and TED-LIUM.
-
Unsupervised Learning of Robust Spectral Shape Matching
Unsupervised deep functional map method with novel loss coupling functional and point-wise maps for robust 3D shape matching on non-isometric, partial, and noisy shapes.
-
dMX: Differentiable Mixed-Precision Assignment for Low-Precision Floating-Point Formats
dMX learns continuous per-layer MXFP bit-width offsets with temperature annealing and target-aware regularization, yielding Pareto-better mixed-precision LLMs than uniform or KL-heuristic baselines.
-
FedLAB: Traceable Semantic Codebooks for Federated Multimodal Graph Foundation Learning
FedLAB organizes multimodal graph knowledge into typed hierarchical codebooks for modality evidence, node semantics, and topology context via federated semantic barycenter pre-training, improving performance by up to 7.53% on benchmarks while enabling semantic traceability.
-
There and Back Again: A Flexible-Frame Transformer for Multi-Exposure Fusion
FreeMEF is the first flexible-frame transformer for multi-exposure fusion using a recurrent state space module and global feature guided block to handle variable numbers of input exposures.
-
Deciphering Fingerprints of 3D Molecular Surfaces for Accurate Epitope Prediction
SurfBind applies a Transformer with patch-level surface modeling and binder-aware cross-attention to 3D molecular surfaces, reporting state-of-the-art epitope prediction on SAbDab and DB5.5 with generalization to unseen antibodies.
-
Joint Discovery of Object and Action Symbols through Effect Prediction for Robotic Manipulation Planning
A binary-bottleneck model discovers object and action symbols from multi-modal effect predictions on random interactions, then uses discrete planning on predicted trajectories for tabletop repositioning and stacking with few-shot generalization to novel objects.
-
Early-Exit Graph Neural Networks for Link Prediction
Early-exit GNNs for link prediction move the speed-quality Pareto frontier on the HeaRT benchmark by allowing implicit early exiting without auxiliary losses.
-
Self-Guidance: Enhancing Neural Codecs via Decoder Manifold Alignment
Self-guidance adds a lightweight feature-mapping loss to align decoder manifolds in VQ-VAE speech codecs, raising reconstruction metrics and allowing 4x codebook reduction with no fidelity loss.
-
Abstract Dynamic Programming on Partially Ordered Spaces
Abstract dynamic programs on partially ordered spaces achieve fundamental optimality properties and convergence of iterations under verifiable topological stability, extending guarantees to undiscounted optimal stopping and Bayesian sequential analysis.
-
Learnable Token Sparsification for Efficient Gigapixel Whole Slide Image Reasoning
Learnable sparsification framework compresses WSI visual tokens to 32 (0.78% of original) via SparseLearn, achieving 73.32% accuracy on SlideBench (TCGA) and outperforming baselines.
-
Beyond Soft Masks: Hard-Perturbation Mixup Explainer for Robust GNN Explainability
HPME proposes hard-perturbation mixup explainer grounded in generalized Graph Information Bottleneck to extract discrete subgraphs and generate in-distribution explanations that outperform soft-mask approaches on synthetic and real datasets.
-
LLM Compression with Jointly Optimizing Architectural and Quantization choices
A differentiable NAS framework jointly optimizes LLM architecture and mixed-precision quantization for linear layers, yielding up to 1.4x faster inference or 6% higher accuracy than sequential baselines on reasoning tasks.
-
Demystifying Multimodal Biomolecular Co-design With Intrinsic Geodesic Coupling
GeoCoupling optimizes temporal couplings between modalities in biomolecular generative models and outperforms synchronous baselines on drug design and protein design tasks.
-
SP-MoMamba: Superpixel-driven Mixture of State Space Experts for Efficient Image Super-Resolution
SP-MoMamba uses superpixels to drive content-aware state space modeling and multi-scale mixture-of-experts for efficient single-image super-resolution.
-
Learning Permutation from Structure Without Supervision
Entropy-adaptive Gumbel-Sinkhorn formulation for unsupervised permutation learning that modulates temperature per assignment to address non-uniform uncertainty.
-
Isomorphic Dynamic Programs
Order isomorphisms transmit optimality across dynamic program formulations, with applications to Epstein-Zin preferences and numerical value function improvements.
-
EvoStruct: Bridging Evolutionary and Structural Priors for Antibody CDR Design via Protein Language Model Adaptation
EvoStruct integrates evolutionary priors from a protein language model with structural priors from an E(3)-equivariant GNN to raise amino acid recovery by 16% and diversity by 2.3x on CHIMERA-Bench while cutting perplexity 43%.
-
ROAR-3D: Routing Arbitrary Views for High-Fidelity 3D Generation
ROAR-3D adds a token-wise view router and dual-stream attention to pretrained single-view 3D generators so they can use arbitrary unposed images for higher-fidelity output.
-
Structured Neural Marked Point Processes for Interpretable Event Interaction Modeling
SNMPP builds a product-form neural influence kernel from a signed interaction network over event classes and a delay-aware monotonic temporal network to enable explicit discovery of inter-event relationships alongside strong prediction.
-
Prognostic Value of Lung Ultrasound Biomarkers for Readmission Risk in Congestive Heart Failure: A Pilot Data-Driven Analysis
Pilot study uses pretrained video encoder features from lung ultrasound to predict 30-day CHF readmission, finding lower-lung views and temporal differences most informative with top MLP F1 of 0.80.