We characterise the problem of abstraction in the context of deep reinforcement learning. Various well established approaches to analogical reasoning and associative memory might be brought to bear on this issue, but they present difficulties because of the need for end-to-end differentiability. We review developments in AI and machine learning that could facilitate their adoption.
Deep reinforcement learning (Deep RL) has recently seen significant progress in developing algorithms for generalization. However, most algorithms target a single type of generalization setting. In this work, we study generalization across three disparate task structures: (a) tasks composed of spatial and temporal compositions of regularly occurring object motions; (b) tasks composed of active perception of and navigation towards regularly occurring 3D objects; and (c) tasks composed of remembering goal-information over sequences of regularly occurring object-configurations. These diverse task structures all share an underlying idea of compositionality: task completion always involves combining recurring segments of task-oriented perception and behavior. We hypothesize that an agent can generalize within a task structure if it can discover representations that capture these recurring task-segments. For our tasks, this corresponds to representations for recognizing individual object motions, for navigation towards 3D objects, and for navigating through object-configurations. Taking inspiration from cognitive science, we term representations for recurring segments of an agent's experience, "perceptual schemas". We propose Feature Attending Recurrent Modules (FARM), which learns a state representation where perceptual schemas are distributed across multiple, relatively small recurrent modules. We compare FARM to recurrent architectures that leverage spatial attention, which reduces observation features to a weighted average over spatial positions. Our experiments indicate that our feature-attention mechanism better enables FARM to generalize across the diverse object-centric domains we study.
We address the problem of building agents whose goal is to satisfy out-of distribution (OOD) multi-task instructions expressed in temporal logic (TL) by using deep reinforcement learning (DRL). Recent works provided evidence that the deep learning architecture is a key feature when teaching a DRL agent to solve OOD tasks in TL. Yet, the studies on their performance are still limited. In this work, we analyse various state-of-the-art (SOTA) architectures that include generalisation mechanisms such as relational layers, the soft-attention mechanism, or hierarchical configurations, when generalising safety-aware tasks expressed in TL. Most importantly, we present a novel deep learning architecture that induces agents to generate latent representations of their current goal given both the human instruction and the current observation from the environment. We find that applying our proposed configuration to SOTA architectures yields significantly stronger performance when executing new tasks in OOD environments.
We present an architecture that is effective for continual learning in an especially demanding setting, where task boundaries do not exist or are unknown. Our architecture comprises an encoder, pre-trained on a separate dataset, and an ensemble of simple one-layer classifiers. Two main innovations are required to make this combination work. First, the provision of suitably generic pre-trained encoders has been made possible thanks to recent progress in self-supervised training methods. Second, pairing each classifier in the ensemble with a key, where the key-space is identical to the latent space of the encoder, allows them to be used collectively, yet selectively, via k-nearest neighbour lookup. We show that models trained with the encoders-and-ensembles architecture are state-of-the-art for the task-free setting on standard image classification continual learning benchmarks, and improve on prior state-of-the-art by a large margin in the most challenging cases. We also show that the architecture learns well in a fully incremental setting, where one class is learned at a time, and we demonstrate its effectiveness in this setting with up to 100 classes. Finally, we show that the architecture works in a task-free continual learning context where the data distribution changes gradually, and existing approaches requiring knowledge of task boundaries cannot be applied.
We present a slot-wise, object-based transition model that decomposes a scene into objects, aligns them (with respect to a slot-wise object memory) to maintain a consistent order across time, and predicts how those objects evolve over successive frames. The model is trained end-to-end without supervision using losses at the level of the object-structured representation rather than pixels. Thanks to its alignment module, the model deals properly with two issues that are not handled satisfactorily by other transition models, namely object persistence and object identity. We show that the combination of an object-level loss and correct object alignment over time enables the model to outperform a state-of-the-art baseline, and allows it to deal well with object occlusion and re-appearance in partially observable environments.
Recently developed deep learning models are able to learn to segment scenes into component objects without supervision. This opens many new and exciting avenues of research, allowing agents to take objects (or entities) as inputs, rather that pixels. Unfortunately, while these models provide excellent segmentation of a single frame, they do not keep track of how objects segmented at one time-step correspond (or align) to those at a later time-step. The alignment (or correspondence) problem has impeded progress towards using object representations in downstream tasks. In this paper we take steps towards solving the alignment problem, presenting the AlignNet, an unsupervised alignment module.
Robust perception relies on both bottom-up and top-down signals. Bottom-up signals consist of what's directly observed through sensation. Top-down signals consist of beliefs and expectations based on past experience and short-term memory, such as how the phrase `peanut butter and~...' will be completed. The optimal combination of bottom-up and top-down information remains an open question, but the manner of combination must be dynamic and both context and task dependent. To effectively utilize the wealth of potential top-down information available, and to prevent the cacophony of intermixed signals in a bidirectional architecture, mechanisms are needed to restrict information flow. We explore deep recurrent neural net architectures in which bottom-up and top-down signals are dynamically combined using attention. Modularity of the architecture further restricts the sharing and communication of information. Together, attention and modularity direct information flow, which leads to reliable performance improvements in perceptual and language tasks, and in particular improves robustness to distractions and noisy data. We demonstrate on a variety of benchmarks in language modeling, sequential image classification, video prediction and reinforcement learning that the \emph{bidirectional} information flow can improve results over strong baselines.
The i.i.d. assumption is a useful idealization that underpins many successful approaches to supervised machine learning. However, its violation can lead to models that learn to exploit spurious correlations in the training data, rendering them vulnerable to adversarial interventions, undermining their reliability, and limiting their practical application. To mitigate this problem, we present a method for learning multiple models, incorporating an objective that pressures each to learn a distinct way to solve the task. We propose a notion of diversity based on minimizing the conditional total correlation of final layer representations across models given the label, which we approximate using a variational estimator and minimize using adversarial training. To demonstrate our framework's ability to facilitate rapid adaptation to distribution shift, we train a number of simple classifiers from scratch on the frozen outputs of our models using a small amount of data from the shifted distribution. Under this evaluation protocol, our framework significantly outperforms a baseline trained using the empirical risk minimization principle.