Tag
This paper introduces sparse priors to improve theoretical guarantees on distribution learning, showing that under a sparse prior, the sample complexity scales as Ω(√(k/n)) instead of O(n^{-1/Θ(d)}), thereby mitigating the curse of dimensionality.
The paper proposes a fast Bayesian optimization method for de novo discovery by leveraging linear models in latent spaces, achieving over 100x speedup over existing methods while maintaining performance.
The paper introduces a flow-matching model for predicting aircraft trajectories using ADS-B data, achieving superior performance over traditional methods in probabilistic trajectory prediction.
The paper presents a unified formulation and empirical comparison of generative-model approaches to nonlinear filtering, deriving new filters based on stochastic interpolants, flow-matching, and Schrödinger bridges, and comparing them against existing methods.
MIT researchers have developed a new method called HardFlow that helps generative AI models satisfy strict safety and task-specific constraints in high-stakes applications by enforcing requirements only on the final output, improving performance in domains like robotics and control systems.
The paper introduces DF26, a benchmark for detecting AI-generated public-speaking videos, showing that both humans and current detectors perform near chance, underscoring the need for improved robustness to generative model distribution shifts.
This paper proposes a four-level framework classifying agentic visual generation systems based on the controller's direct decision scope over generation operations, from fixed support to experience-adaptive control.
LeanGRPO eliminates redundant recomputation in diffusion RL methods by introducing recompute-free training schedules, achieving up to 1.83x speedup while preserving optimization objectives.
This paper investigates the link between autoencoder parameters and data statistics, proposing that parameters can function as a vector representation of data, supported by theoretical analysis and experiments on CIFAR-10 and FashionMNIST.
This paper introduces Quantile AlignTree Flow Matching (QAT-FM), a structured coupling method for flow matching that constructs hierarchical couplings using quantile-aligned trees to achieve non-crossing paths and efficient training for high-dimensional generative tasks.
This paper demonstrates that causal fairness mechanisms, specifically edge cuts on causal graphs, are portable across various synthetic data generator families including GANs and diffusion models, with minimal impact on data fidelity and utility.
The paper investigates the conditional diversity gap in large language models by comparing the entropy of generated outputs with training data and proposes an information-theoretic framework to measure and mitigate this gap.
The paper introduces Elite-Weighted Supervised Fine-tuning (EW-SFT), a method for goal-directed molecular optimization that uses reward to guide elite selection and updates via the model's native loss, applicable across various generative architectures and tasks.
This paper investigates the effects of inference-time interventions and weight consolidation on open-ended AI generation, finding that training on value-filtered candidates improves mean quality but does not exceed classic heuristic performance.
This article provides an overview of 20 key machine learning models that underpin modern AI systems, explaining their roles in tasks from prediction to generative art.
This paper introduces Gromov-Monge flow matching for equivariant graph generation, improving sample quality with structure-aware couplings compatible with standard architectures.
The paper proposes a self-augmented diffusion guidance method to incorporate physical laws into diffusion models, reducing deviations from true dynamics and enabling faster generation.
GRAS is a training-free method for reward alignment in discrete diffusion models that reduces variance in guided proposals and adaptively selects particles, achieving state-of-the-art results on DNA and protein design tasks.
Magpie is a real-time generative rendering system for interactive games that separates gameplay logic from visual generation to reduce asset production costs and preserve interactive designability.
The article introduces ActFlow, a continued pre-training scheme that expands the valid design space for flow and diffusion models, enabling out-of-distribution generative modeling and evolvable search spaces in scientific discovery.