Reinforcement studying (RL) algorithms can study expertise to resolve decision-making duties like enjoying video games, enabling robots to choose up objects, and even optimizing microchip designs. Nonetheless, operating RL algorithms in the actual world requires costly energetic knowledge assortment. Pre-training on various datasets has confirmed to allow data-efficient fine-tuning for particular person downstream duties in pure language processing (NLP) and imaginative and prescient issues. In the identical method that BERT or GPT-3 fashions present general-purpose initialization for NLP, massive RL–pre-trained fashions may present general-purpose initialization for decision-making. So, we ask the query: Can we allow related pre-training to speed up RL strategies and create a general-purpose “spine” for environment friendly RL throughout numerous duties?
In “Offline Q-learning on Numerous Multi-Job Information Each Scales and Generalizes”, to be revealed at ICLR 2023, we focus on how we scaled offline RL, which can be utilized to coach worth features on beforehand collected static datasets, to offer such a normal pre-training technique. We show that Scaled Q-Studying utilizing a various dataset is enough to study representations that facilitate fast switch to novel duties and quick on-line studying on new variations of a job, bettering considerably over present illustration studying approaches and even Transformer-based strategies that use a lot bigger fashions.
Scaled Q-learning: Multi-task pre-training with conservative Q-learning
To offer a general-purpose pre-training strategy, offline RL must be scalable, permitting us to pre-train on knowledge throughout totally different duties and make the most of expressive neural community fashions to amass highly effective pre-trained backbones, specialised to particular person downstream duties. We based mostly our offline RL pre-training technique on conservative Q-learning (CQL), a easy offline RL technique that mixes normal Q-learning updates with a further regularizer that minimizes the worth of unseen actions. With discrete actions, the CQL regularizer is equal to a regular cross-entropy loss, which is an easy, one-line modification on normal deep Q-learning. A couple of essential design choices made this doable:
Neural community measurement: We discovered that multi-game Q-learning required massive neural community architectures. Whereas prior strategies typically used comparatively shallow convolutional networks, we discovered that fashions as massive as a ResNet 101 led to vital enhancements over smaller fashions.
Neural community structure: To study pre-trained backbones which can be helpful for brand new video games, our last structure makes use of a shared neural community spine, with separate 1-layer heads outputting Q-values of every recreation. This design avoids interference between the video games throughout pre-training, whereas nonetheless offering sufficient knowledge sharing to study a single shared illustration. Our shared imaginative and prescient spine additionally utilized a realized place embedding (akin to Transformer fashions) to maintain observe of spatial data within the recreation.
Representational regularization: Current work has noticed that Q-learning tends to undergo from representational collapse points, the place even massive neural networks can fail to study efficient representations. To counteract this concern, we leverage our prior work to normalize the final layer options of the shared a part of the Q-network. Moreover, we utilized a categorical distributional RL loss for Q-learning, which is thought to offer richer representations that enhance downstream job efficiency.
The multi-task Atari benchmark
We consider our strategy for scalable offline RL on a set of Atari video games, the place the aim is to coach a single RL agent to play a set of video games utilizing heterogeneous knowledge from low-quality (i.e., suboptimal) gamers, after which use the ensuing community spine to rapidly study new variations in pre-training video games or utterly new video games. Coaching a single coverage that may play many various Atari video games is troublesome sufficient even with normal on-line deep RL strategies, as every recreation requires a distinct technique and totally different representations. Within the offline setting, some prior works, corresponding to multi-game resolution transformers, proposed to dispense with RL totally, and as an alternative make the most of conditional imitation studying in an try and scale with massive neural community architectures, corresponding to transformers. Nonetheless, on this work, we present that this type of multi-game pre-training will be performed successfully by way of RL by using CQL together with just a few cautious design choices, which we describe under.
Scalability on coaching video games
We consider the Scaled Q-Studying technique’s efficiency and scalability utilizing two knowledge compositions: (1) close to optimum knowledge, consisting of all of the coaching knowledge showing in replay buffers of earlier RL runs, and (2) low high quality knowledge, consisting of knowledge from the primary 20% of the trials within the replay buffer (i.e., solely knowledge from extremely suboptimal insurance policies). In our outcomes under, we evaluate Scaled Q-Studying with an 80-million parameter mannequin to multi-game resolution transformers (DT) with both 40-million or 80-million parameter fashions, and a behavioral cloning (imitation studying) baseline (BC). We observe that Scaled Q-Studying is the one strategy that improves over the offline knowledge, attaining about 80% of human normalized efficiency.
Additional, as proven under, Scaled Q-Studying improves when it comes to efficiency, however it additionally enjoys favorable scaling properties: simply as how the efficiency of pre-trained language and imaginative and prescient fashions improves as community sizes get larger, having fun with what is usually referred as “power-law scaling”, we present that the efficiency of Scaled Q-learning enjoys related scaling properties. Whereas this can be unsurprising, this type of scaling has been elusive in RL, with efficiency typically deteriorating with bigger mannequin sizes. This means that Scaled Q-Studying together with the above design selections higher unlocks the flexibility of offline RL to make the most of massive fashions.
Advantageous-tuning to new video games and variations
To guage fine-tuning from this offline initialization, we think about two settings: (1) fine-tuning to a brand new, totally unseen recreation with a small quantity of offline knowledge from that recreation, comparable to 2M transitions of gameplay, and (2) fine-tuning to a brand new variant of the video games with on-line interplay. The fine-tuning from offline gameplay knowledge is illustrated under. Observe that this situation is mostly extra favorable to imitation-style strategies, Determination Transformer and behavioral cloning, for the reason that offline knowledge for the brand new video games is of comparatively high-quality. Nonetheless, we see that most often Scaled Q-learning improves over different approaches (80% on common), in addition to devoted illustration studying strategies, corresponding to MAE or CPC, which solely use the offline knowledge to study visible representations slightly than worth features.
Within the on-line setting, we see even bigger enhancements from pre-training with Scaled Q-learning. On this case, illustration studying strategies like MAE yield minimal enchancment throughout on-line RL, whereas Scaled Q-Studying can efficiently combine prior information concerning the pre-training video games to considerably enhance the ultimate rating after 20k on-line interplay steps.
These outcomes show that pre-training generalist worth perform backbones with multi-task offline RL can considerably increase efficiency of RL on downstream duties, each in offline and on-line mode. Observe that these fine-tuning duties are fairly troublesome: the assorted Atari video games, and even variants of the identical recreation, differ considerably in look and dynamics. For instance, the goal blocks in Breakout disappear within the variation of the sport as proven under, making management troublesome. Nonetheless, the success of Scaled Q-learning, notably as in comparison with visible illustration studying methods, corresponding to MAE and CPC, means that the mannequin is in actual fact studying some illustration of the sport dynamics, slightly than merely offering higher visible options.
Advantageous-tuning with on-line RL for variants of the sport Freeway, Hero, and Breakout. The brand new variant utilized in fine-tuning is proven within the backside row of every determine, the unique recreation seen in pre-training is within the prime row. Advantageous-tuning from Scaled Q-Studying considerably outperforms MAE (a visible illustration studying technique) and studying from scratch with single-game DQN.
Conclusion and takeaways
We offered Scaled Q-Studying, a pre-training technique for scaled offline RL that builds on the CQL algorithm, and demonstrated the way it allows environment friendly offline RL for multi-task coaching. This work made preliminary progress in direction of enabling extra sensible real-world coaching of RL brokers as a substitute for pricey and complicated simulation-based pipelines or large-scale experiments. Maybe in the long term, related work will result in usually succesful pre-trained RL brokers that develop broadly relevant exploration and interplay expertise from large-scale offline pre-training. Validating these outcomes on a broader vary of extra sensible duties, in domains corresponding to robotics (see some preliminary outcomes) and NLP, is a vital path for future analysis. Offline RL pre-training has quite a lot of potential, and we count on that we are going to see many advances on this space in future work.
Acknowledgements
This work was performed by Aviral Kumar, Rishabh Agarwal, Xinyang Geng, George Tucker, and Sergey Levine. Particular due to Sherry Yang, Ofir Nachum, and Kuang-Huei Lee for assist with the multi-game resolution transformer codebase for analysis and the multi-game Atari benchmark, and Tom Small for illustrations and animation.
