This work introduces Gemma, a family of lightweight, state-of-the art open models built from the research and technology used to create Gemini models. Gemma models demonstrate strong performance across academic benchmarks for language understanding, reasoning, and safety. We release two sizes of models (2 billion and 7 billion parameters), and provide both pretrained and fine-tuned checkpoints. Gemma outperforms similarly sized open models on 11 out of 18 text-based tasks, and we present comprehensive evaluations of safety and responsibility aspects of the models, alongside a detailed description of model development. We believe the responsible release of LLMs is critical for improving the safety of frontier models, and for enabling the next wave of LLM innovations.
Standard approaches to sequential decision-making exploit an agent's ability to continually interact with its environment and improve its control policy. However, due to safety, ethical, and practicality constraints, this type of trial-and-error experimentation is often infeasible in many real-world domains such as healthcare and robotics. Instead, control policies in these domains are typically trained offline from previously logged data or in a growing-batch manner. In this setting a fixed policy is deployed to the environment and used to gather an entire batch of new data before being aggregated with past batches and used to update the policy. This improvement cycle can then be repeated multiple times. While a limited number of such cycles is feasible in real-world domains, the quality and diversity of the resulting data are much lower than in the standard continually-interacting approach. However, data collection in these domains is often performed in conjunction with human experts, who are able to label or annotate the collected data. In this paper, we first explore the trade-offs present in this growing-batch setting, and then investigate how information provided by a teacher (i.e., demonstrations, expert actions, and gradient information) can be leveraged at training time to mitigate the sample complexity and coverage requirements for actor-critic methods. We validate our contributions on tasks from the DeepMind Control Suite.
Actor-critic algorithms that make use of distributional policy evaluation have frequently been shown to outperform their non-distributional counterparts on many challenging control tasks. Examples of this behavior include the D4PG and DMPO algorithms as compared to DDPG and MPO, respectively [Barth-Maron et al., 2018; Hoffman et al., 2020]. However, both agents rely on the C51 critic for value estimation.One major drawback of the C51 approach is its requirement of prior knowledge about the minimum andmaximum values a policy can attain as well as the number of bins used, which fixes the resolution ofthe distributional estimate. While the DeepMind control suite of tasks utilizes standardized rewards and episode lengths, thus enabling the entire suite to be solved with a single setting of these hyperparameters, this is often not the case. This paper revisits a natural alternative that removes this requirement, namelya mixture of Gaussians, and a simple sample-based loss function to train it in an off-policy regime. We empirically evaluate its performance on a broad range of continuous control tasks and demonstrate that it eliminates the need for these distributional hyperparameters and achieves state-of-the-art performance on a variety of challenging tasks (e.g. the humanoid, dog, quadruped, and manipulator domains). Finallywe provide an implementation in the Acme agent repository.
Actor-critic algorithms that make use of distributional policy evaluation have frequently been shown to outperform their non-distributional counterparts on many challenging control tasks. Examples of this behavior include the D4PG and DMPO algorithms as compared to DDPG and MPO, respectively [Barth-Maron et al., 2018; Hoffman et al., 2020]. However, both agents rely on the C51 critic for value estimation.One major drawback of the C51 approach is its requirement of prior knowledge about the minimum andmaximum values a policy can attain as well as the number of bins used, which fixes the resolution ofthe distributional estimate. While the DeepMind control suite of tasks utilizes standardized rewards and episode lengths, thus enabling the entire suite to be solved with a single setting of these hyperparameters, this is often not the case. This paper revisits a natural alternative that removes this requirement, namelya mixture of Gaussians, and a simple sample-based loss function to train it in an off-policy regime. We empirically evaluate its performance on a broad range of continuous control tasks and demonstrate that it eliminates the need for these distributional hyperparameters and achieves state-of-the-art performance on a variety of challenging tasks (e.g. the humanoid, dog, quadruped, and manipulator domains). Finallywe provide an implementation in the Acme agent repository.
Many advances that have improved the robustness and efficiency of deep reinforcement learning (RL) algorithms can, in one way or another, be understood as introducing additional objectives, or constraints, in the policy optimization step. This includes ideas as far ranging as exploration bonuses, entropy regularization, and regularization toward teachers or data priors when learning from experts or in offline RL. Often, task reward and auxiliary objectives are in conflict with each other and it is therefore natural to treat these examples as instances of multi-objective (MO) optimization problems. We study the principles underlying MORL and introduce a new algorithm, Distillation of a Mixture of Experts (DiME), that is intuitive and scale-invariant under some conditions. We highlight its strengths on standard MO benchmark problems and consider case studies in which we recast offline RL and learning from experts as MO problems. This leads to a natural algorithmic formulation that sheds light on the connection between existing approaches. For offline RL, we use the MO perspective to derive a simple algorithm, that optimizes for the standard RL objective plus a behavioral cloning term. This outperforms state-of-the-art on two established offline RL benchmarks.
Offline reinforcement learning (RL), also known as batch RL, offers the prospect of policy optimization from large pre-recorded datasets without online environment interaction. It addresses challenges with regard to the cost of data collection and safety, both of which are particularly pertinent to real-world applications of RL. Unfortunately, most off-policy algorithms perform poorly when learning from a fixed dataset. In this paper, we propose a novel offline RL algorithm to learn policies from data using a form of critic-regularized regression (CRR). We find that CRR performs surprisingly well and scales to tasks with high-dimensional state and action spaces -- outperforming several state-of-the-art offline RL algorithms by a significant margin on a wide range of benchmark tasks.
Deep reinforcement learning has led to many recent-and groundbreaking-advancements. However, these advances have often come at the cost of both the scale and complexity of the underlying RL algorithms. Increases in complexity have in turn made it more difficult for researchers to reproduce published RL algorithms or rapidly prototype ideas. To address this, we introduce Acme, a tool to simplify the development of novel RL algorithms that is specifically designed to enable simple agent implementations that can be run at various scales of execution. Our aim is also to make the results of various RL algorithms developed in academia and industrial labs easier to reproduce and extend. To this end we are releasing baseline implementations of various algorithms, created using our framework. In this work we introduce the major design decisions behind Acme and show how these are used to construct these baselines. We also experiment with these agents at different scales of both complexity and computation-including distributed versions. Ultimately, we show that the design decisions behind Acme lead to agents that can be scaled both up and down and that, for the most part, greater levels of parallelization result in agents with equivalent performance, just faster.
This paper introduces R2D3, an agent that makes efficient use of demonstrations to solve hard exploration problems in partially observable environments with highly variable initial conditions. We also introduce a suite of eight tasks that combine these three properties, and show that R2D3 can solve several of the tasks where other state of the art methods (both with and without demonstrations) fail to see even a single successful trajectory after tens of billions of steps of exploration.
We propose a novel framework for the analysis of learning algorithms that allows us to say when such algorithms can and cannot generalize certain patterns from training data to test data. In particular we focus on situations where the rule that must be learned concerns two components of a stimulus being identical. We call such a basis for discrimination an identity-based rule. Identity-based rules have proven to be difficult or impossible for certain types of learning algorithms to acquire from limited datasets. This is in contrast to human behaviour on similar tasks. Here we provide a framework for rigorously establishing which learning algorithms will fail at generalizing identity-based rules to novel stimuli. We use this framework to show that such algorithms are unable to generalize identity-based rules to novel inputs unless trained on virtually all possible inputs. We demonstrate these results computationally with a multilayer feedforward neural network.
Bayesian optimization has recently emerged as a popular and efficient tool for global optimization and hyperparameter tuning. Currently, the established Bayesian optimization practice requires a user-defined bounding box which is assumed to contain the optimizer. However, when little is known about the probed objective function, it can be difficult to prescribe such bounds. In this work we modify the standard Bayesian optimization framework in a principled way to allow automatic resizing of the search space. We introduce two alternative methods and compare them on two common synthetic benchmarking test functions as well as the tasks of tuning the stochastic gradient descent optimizer of a multi-layered perceptron and a convolutional neural network on MNIST.