One of the major drawbacks of deep learning models for computer vision has been their inability to retain multiple sources of information in a modular fashion. For instance, given a network that has been trained on a source task, we would like to re-train this network on a similar, yet different, target task while maintaining its performance on the source task. Simultaneously, researchers have extensively studied modularization of network weights to localize and identify the set of weights culpable for eliciting the observed performance on a given task. One set of works studies the modularization induced in the weights of a neural network by learning and analysing weight masks. In this work, we combine these fields to study three such weight masking methods and analyse their ability to mitigate "forgetting'' on the source task while also allowing for efficient finetuning on the target task. We find that different masking techniques have trade-offs in retaining knowledge in the source task without adversely affecting target task performance.

Reinforcement learning is the classic paradigm for addressing sequential decision-making problems [47]. Naturally, while inheriting the fundamental challenge of generalization across novel states and actions from supervised learning, general-purpose reinforcement-learning agents must also contend with the additional challenges of exploration and credit assignment. While much initial progress in the field was driven largely by innovative machinery for tackling credit assignment [45, 46, 44] alongside simple exploration heuristics (ε-greedy exploration, for example), recent years have seen a reversal with the bulk of attention focused on a broad array of exploration methods (spanning additional heuristics as well as more principled approaches) [51, 38, 15], and relatively little consideration given to issues of credit assignment. This lack of interest in solution concepts, however, has not stopped the proliferation of reinforcement learning into novel application areas characterized by long problem horizons and sparse reward signals; indeed, the current reinforcement learning from human feedback (RLHF) paradigm [28] is now a widely popularized example of an environment that operates in perhaps the harshest setting where a single feedback signal is only obtained after the completion of a long trajectory.

A key challenge for AI is to build embodied systems that operate in dynamically changing environments. Such systems must adapt to changing task contexts and learn continuously. Although standard deep learning systems achieve state of the art results on static benchmarks, they often struggle in dynamic scenarios. In these settings, error signals from multiple contexts can interfere with one another, ultimately leading to a phenomenon known as catastrophic forgetting. In this article we investigate biologically inspired architectures as solutions to these problems. Specifically, we show that the biophysical properties of dendrites and local inhibitory systems enable networks to dynamically restrict and route information in a context-specific manner. Our key contributions are as follows. First, we propose a novel artificial neural network architecture that incorporates active dendrites and sparse representations into the standard deep learning framework. Next, we study the performance of this architecture on two separate benchmarks requiring task-based adaptation: Meta-World, a multi-task reinforcement learning environment where a robotic agent must learn to solve a variety of manipulation tasks simultaneously; and a continual learning benchmark in which the model's prediction task changes throughout training. Analysis on both benchmarks demonstrates the emergence of overlapping but distinct and sparse subnetworks, allowing the system to fluidly learn multiple tasks with minimal forgetting. Our neural implementation marks the first time a single architecture has achieved competitive results on both multi-task and continual learning settings. Our research sheds light on how biological properties of neurons can inform deep learning systems to address dynamic scenarios that are typically impossible for traditional ANNs to solve.

Proximal Policy Optimization (PPO) is a popular on-policy reinforcement learning algorithm but is significantly less utilized than off-policy learning algorithms in multi-agent problems. In this work, we investigate Multi-Agent PPO (MAPPO), a multi-agent PPO variant which adopts a centralized value function. Using a 1-GPU desktop, we show that MAPPO achieves performance comparable to the state-of-the-art in three popular multi-agent testbeds: the Particle World environments, Starcraft II Micromanagement Tasks, and the Hanabi Challenge, with minimal hyperparameter tuning and without any domain-specific algorithmic modifications or architectures. In the majority of environments, we find that compared to off-policy baselines, MAPPO achieves better or comparable sample complexity as well as substantially faster running time. Finally, we present 5 factors most influential to MAPPO's practical performance with ablation studies.