Elucidating Optimal Reward-Diversity Tradeoffs in Text-to-Image Diffusion Models

Text-to-image (T2I) diffusion models have become prominent tools for generating high-fidelity images from text prompts. However, when trained on unfiltered internet data, these models can produce unsafe, incorrect, or stylistically undesirable images that are not aligned with human preferences. To address this, recent approaches have incorporated human preference datasets to fine-tune T2I models or to optimize reward functions that capture these preferences. Although effective, these methods are vulnerable to reward hacking, where the model overfits to the reward function, leading to a loss of diversity in the generated images. In this paper, we prove the inevitability of reward hacking and study natural regularization techniques like KL divergence and LoRA scaling, and their limitations for diffusion models. We also introduce Annealed Importance Guidance (AIG), an inference-time regularization inspired by Annealed Importance Sampling, which retains the diversity of the base model while achieving Pareto-Optimal reward-diversity tradeoffs. Our experiments demonstrate the benefits of AIG for Stable Diffusion models, striking the optimal balance between reward optimization and image diversity. Furthermore, a user study confirms that AIG improves diversity and quality of generated images across different model architectures and reward functions.

NeMo-Aligner: Scalable Toolkit for Efficient Model Alignment

Aligning Large Language Models (LLMs) with human values and preferences is essential for making them helpful and safe. However, building efficient tools to perform alignment can be challenging, especially for the largest and most competent LLMs which often contain tens or hundreds of billions of parameters. We create NeMo-Aligner, a toolkit for model alignment that can efficiently scale to using hundreds of GPUs for training. NeMo-Aligner comes with highly optimized and scalable implementations for major paradigms of model alignment such as: Reinforcement Learning from Human Feedback (RLHF), Direct Preference Optimization (DPO), SteerLM, and Self-Play Fine-Tuning (SPIN). Additionally, our toolkit supports running most of the alignment techniques in a Parameter Efficient Fine-Tuning (PEFT) setting. NeMo-Aligner is designed for extensibility, allowing support for other alignment techniques with minimal effort. It is open-sourced with Apache 2.0 License and we invite community contributions at https://github.com/NVIDIA/NeMo-Aligner

Hierarchical Graph Neural Network with Cross-Attention for Cross-Device User Matching

Cross-device user matching is a critical problem in numerous domains, including advertising, recommender systems, and cybersecurity. It involves identifying and linking different devices belonging to the same person, utilizing sequence logs. Previous data mining techniques have struggled to address the long-range dependencies and higher-order connections between the logs. Recently, researchers have modeled this problem as a graph problem and proposed a two-tier graph contextual embedding (TGCE) neural network architecture, which outperforms previous methods. In this paper, we propose a novel hierarchical graph neural network architecture (HGNN), which has a more computationally efficient second level design than TGCE. Furthermore, we introduce a cross-attention (Cross-Att) mechanism in our model, which improves performance by 5% compared to the state-of-the-art TGCE method.

MG-GNN: Multigrid Graph Neural Networks for Learning Multilevel Domain Decomposition Methods

Domain decomposition methods (DDMs) are popular solvers for discretized systems of partial differential equations (PDEs), with one-level and multilevel variants. These solvers rely on several algorithmic and mathematical parameters, prescribing overlap, subdomain boundary conditions, and other properties of the DDM. While some work has been done on optimizing these parameters, it has mostly focused on the one-level setting or special cases such as structured-grid discretizations with regular subdomain construction. In this paper, we propose multigrid graph neural networks (MG-GNN), a novel GNN architecture for learning optimized parameters in two-level DDMs\@. We train MG-GNN using a new unsupervised loss function, enabling effective training on small problems that yields robust performance on unstructured grids that are orders of magnitude larger than those in the training set. We show that MG-GNN outperforms popular hierarchical graph network architectures for this optimization and that our proposed loss function is critical to achieving this improved performance.

Optimized Sparse Matrix Operations for Reverse Mode Automatic Differentiation

Sparse matrix representations are ubiquitous in computational science and machine learning, leading to significant reductions in compute time, in comparison to dense representation, for problems that have local connectivity. The adoption of sparse representation in leading ML frameworks such as PyTorch is incomplete, however, with support for both automatic differentiation and GPU acceleration missing. In this work, we present an implementation of a CSR-based sparse matrix wrapper for PyTorch with CUDA acceleration for basic matrix operations, as well as automatic differentiability. We also present several applications of the resulting sparse kernels to optimization problems, demonstrating ease of implementation and performance measurements versus their dense counterparts.

Learning Interface Conditions in Domain Decomposition Solvers

Domain decomposition methods are widely used and effective in the approximation of solutions to partial differential equations. Yet the optimal construction of these methods requires tedious analysis and is often available only in simplified, structured-grid settings, limiting their use for more complex problems. In this work, we generalize optimized Schwarz domain decomposition methods to unstructured-grid problems, using Graph Convolutional Neural Networks (GCNNs) and unsupervised learning to learn optimal modifications at subdomain interfaces. A key ingredient in our approach is an improved loss function, enabling effective training on relatively small problems, but robust performance on arbitrarily large problems, with computational cost linear in problem size. The performance of the learned linear solvers is compared with both classical and optimized domain decomposition algorithms, for both structured- and unstructured-grid problems.

Optimization-Based Algebraic Multigrid Coarsening Using Reinforcement Learning

Large sparse linear systems of equations are ubiquitous in science and engineering, such as those arising from discretizations of partial differential equations. Algebraic multigrid (AMG) methods are one of the most common methods of solving such linear systems, with an extensive body of underlying mathematical theory. A system of linear equations defines a graph on the set of unknowns and each level of a multigrid solver requires the selection of an appropriate coarse graph along with restriction and interpolation operators that map to and from the coarse representation. The efficiency of the multigrid solver depends critically on this selection and many selection methods have been developed over the years. Recently, it has been demonstrated that it is possible to directly learn the AMG interpolation and restriction operators, given a coarse graph selection. In this paper, we consider the complementary problem of learning to coarsen graphs for a multigrid solver. We propose a method using a reinforcement learning (RL) agent based on graph neural networks (GNNs), which can learn to perform graph coarsening on small training graphs and then be applied to unstructured large graphs. We demonstrate that this method can produce better coarse graphs than existing algorithms, even as the graph size increases and other properties of the graph are varied. We also propose an efficient inference procedure for performing graph coarsening that results in linear time complexity in graph size.

Local Navigation and Docking of an Autonomous Robot Mower using Reinforcement Learning and Computer Vision

We demonstrate a successful navigation and docking control system for the John Deere Tango autonomous mower, using only a single camera as the input. This vision-only system is of interest because it is inexpensive, simple for production, and requires no external sensing. This is in contrast to existing systems that rely on integrated position sensors and global positioning system (GPS) technologies. To produce our system we combined a state-of-the-art object detection architecture, YOLO, with a reinforcement learning (RL) architecture, Double Deep QNetworks (Double DQN). The object detection network identifies features on the mower and passes its output to the RL network, providing it with a low-dimensional representation that enables rapid and robust training. Finally, the RL network learns how to navigate the machine to the desired spot in a custom simulation environment. When tested on mower hardware the system is able to dock with centimeter-level accuracy from arbitrary initial locations and orientations.