The Statistical Structure of Identifiable Generative Models
Explore the statistical structure of identifiable generative models, focusing on causal representation learning, model identifiability, and unique inferences from observations under suitable assumptions.
Action Matching: Learning Stochastic Dynamics from Samples
Explore Action Matching for learning stochastic dynamics from samples. Discover applications in quantum systems, biology, and generative modeling. Gain insights into novel approaches for simulating trajectories.
Equivariant Flow Matching for Deep Generative Models in Physics
Explore equivariant flow matching for efficient, simulation-free training of equivariant CNFs in physics and drug discovery, improving sampling efficiency and scalability for many-particle systems.
PFGM++: Unlocking the Potential of Physics-Inspired Generative Models
Explore PFGM++, a physics-inspired generative model unifying diffusion models and PFGMs. Learn about its flexible embedding, unbiased objectives, and superior performance on image datasets.
Improving and Generalizing Flow-Based Generative Models with Minibatch Optimal Transport
Explore advanced techniques for improving continuous normalizing flows, including generalized conditional flow matching and optimal transport CFM, enhancing generative modeling for various applications.
Generative Language Models in Molecular Discovery: Regression Transformer, GT4SD and Beyond
Explore generative language models for molecular discovery, focusing on the Regression Transformer and GT4SD toolkit. Learn about multitask approaches bridging regression and conditional generation for scientific applications.
Clifford Group Equivariant Neural Networks
Explore Clifford Group Equivariant Neural Networks for E(n)-equivariant network construction. Discover theoretical foundations, methodology, and applications in various dimensions and tasks.
Particle Guidance: Non-I.I.D. Diverse Sampling with Diffusion Models
Explore particle guidance for diverse sampling in diffusion models, enhancing diversity and efficiency in generative tasks like image and molecular conformer generation.
Mirror Diffusion Models for Constrained and Watermarked Generation
Explore Mirror Diffusion Models for generating data on constrained sets, enhancing watermarking capabilities, and improving AI safety in drug discovery applications.
Designing Losses for Data-Free Training of Normalizing Flows on Boltzmann Distributions
Explore advanced techniques for training normalizing flows on Boltzmann distributions without data, addressing mode collapse and optimizing high-dimensional distributions in molecular configuration generation.
Stochastic Interpolants: A Unifying Framework for Flows and Diffusions - Lecture
Explore a unified framework for generative models, bridging flow-based and diffusion-based methods using stochastic interpolants. Learn about probability flow equations, likelihood estimation, and applications in AI for drug discovery.
Register Tiling for Unstructured Sparsity in Neural Network Inference
Explore a novel technique for optimizing sparse matrix multiplication in neural networks, improving performance for unstructured sparsity patterns in ML models on CPUs and edge devices.
Scallop: A Language for Neurosymbolic Programming
Explore Scallop, a language combining deep learning and logical reasoning for efficient neurosymbolic programming. Learn its key features and applications in diverse AI tasks.
Offloading-Efficient Sparse AI Systems
Explore offloading-efficient sparse AI systems to optimize performance and resource utilization in artificial intelligence applications.
Tackling the Challenges of High-Performance Graph Analytics at Compiler Level
Explore COMET, a DSL and compiler framework for efficient tensor algebra code generation, focusing on optimizations for high-performance graph analytics on heterogeneous systems.
Challenges and Opportunities for Sparse Compilers in Large Language Models
Explore challenges and opportunities for sparse compilers in large language models, focusing on advancements in compiler technology for AI applications.
Predicting Dynamic Properties of Heap Allocations using Neural Networks Trained on Static Code
Explore neural network models predicting heap allocation properties from static code, examining trade-offs, promising directions, and challenges in this innovative approach to memory management.
Software and Hardware for Sparse Machine Learning
Explore cutting-edge software and hardware solutions for sparse machine learning, enhancing efficiency and performance in ML applications.
Inductive Program Synthesis via Iterative Forward-Backward Abstract Interpretation
Explore a novel approach to inductive program synthesis using iterative forward-backward abstract interpretation, enhancing search space pruning and improving synthesis efficiency.
One Pixel Adversarial Attacks via Sketched Programs
Explore a novel approach for one-pixel adversarial attacks using program synthesis, achieving state-of-the-art success rates with fewer queries and transferable programs for neural network classifiers.
Trace-Guided Inductive Synthesis of Recursive Functional Programs
Explore a novel trace-guided approach for synthesizing recursive functional programs from input-output examples, addressing ambiguity and generalization challenges in program synthesis.