Metacognition, defined as the awareness and regulation of one's cognitive processes, is central to human adaptability in unknown situations. In contrast, current autonomous agents often struggle in novel environments due to their limited capacity for adaptation. We hypothesize that metacognition is a critical missing ingredient in autonomous agents for the cognitive flexibility needed to tackle unfamiliar challenges. Given the broad scope of metacognitive abilities, we focus on competence awareness and strategy selection. To this end, we propose the Metacognition for Unknown Situations and Environments (MUSE) framework to integrate metacognitive processes of self-assessment and self-regulation into autonomous agents. We present two implementations of MUSE: one based on world modeling and another leveraging large language models (LLMs). Our system continually learns to assess its competence on a given task and uses this self-assessment to guide iterative cycles of strategy selection. MUSE agents demonstrate high competence awareness and significant improvements in self-regulation for solving novel, out-of-distribution tasks more effectively compared to model-based reinforcement learning and purely prompt-based LLM agent approaches. This work highlights the promise of approaches inspired by cognitive and neural systems in enabling autonomous agents to adapt to new environments while mitigating the heavy reliance on extensive training data and large models for the current models.

Meta-reinforcement learning (meta-RL) algorithms enable agents to adapt quickly to tasks from few samples in dynamic environments. Such a feat is achieved through dynamic representations in an agent's policy network (obtained via reasoning about task context, model parameter updates, or both). However, obtaining rich dynamic representations for fast adaptation beyond simple benchmark problems is challenging due to the burden placed on the policy network to accommodate different policies. This paper addresses the challenge by introducing neuromodulation as a modular component to augment a standard policy network that regulates neuronal activities in order to produce efficient dynamic representations for task adaptation. The proposed extension to the policy network is evaluated across multiple discrete and continuous control environments of increasing complexity. To prove the generality and benefits of the extension in meta-RL, the neuromodulated network was applied to two state-of-the-art meta-RL algorithms (CAVIA and PEARL). The result demonstrates that meta-RL augmented with neuromodulation produces significantly better result and richer dynamic representations in comparison to the baselines.

In this article, we consider a subclass of partially observable Markov decision process (POMDP) problems which we termed confounding POMDPs. In these types of POMDPs, temporal difference (TD)-based reinforcement learning (RL) algorithms struggle, as TD error cannot be easily derived from observations. We solve these types of problems using a new bio-inspired neural architecture that combines a modulated Hebbian network (MOHN) with deep Q-network (DQN), which we call modulated Hebbian plus Q-network architecture (MOHQA). The key idea is to use a Hebbian network with rarely correlated bio-inspired neural traces to bridge temporal delays between actions and rewards when confounding observations and sparse rewards result in inaccurate TD errors. In MOHQA, DQN learns low-level features and control, while the MOHN contributes to high-level decisions by associating rewards with past states and actions. Thus, the proposed architecture combines two modules with significantly different learning algorithms, a Hebbian associative network and a classical DQN pipeline, exploiting the advantages of both. Simulations on a set of POMDPs and on the Malmo environment show that the proposed algorithm improved DQN's results and even outperformed control tests with advantage-actor critic (A2C), quantile regression DQN with long short-term memory (QRDQN + LSTM), Monte Carlo policy gradient (REINFORCE), and aggregated memory for reinforcement learning (AMRL) algorithms on most difficult POMDPs with confounding stimuli and sparse rewards.

Preventing catastrophic forgetting while continually learning new tasks is an essential problem in lifelong learning. Structural regularization (SR) refers to a family of algorithms that mitigate catastrophic forgetting by penalizing the network for changing its “critical parameters" from previous tasks while learning a new one. The penalty is often induced via a quadratic regularizer defined by an \emph{importance matrix}, e.g., the (empirical) Fisher information matrix in the Elastic Weight Consolidation framework. In practice and due to computational constraints, most SR methods crudely approximate the importance matrix by its diagonal. In this paper, we propose \emph{Sketched Structural Regularization} (Sketched SR) as an alternative approach to compress the importance matrices used for regularizing in SR methods. Specifically, we apply \emph{linear sketching methods} to better approximate the importance matrices in SR algorithms. We show that sketched SR: (i) is computationally efficient and straightforward to implement, (ii) provides an approximation error that is justified in theory, and (iii) is method oblivious by construction and can be adapted to any method that belongs to the SR class. We show that our proposed approach consistently improves various SR algorithms’ performance on both synthetic experiments and benchmark continual learning tasks, including permuted-MNIST and CIFAR-100.

After learning a concept, humans are also able to continually generalize their learned concepts to new domains by observing only a few labeled instances without any interference with the past learned knowledge. In contrast, learning concepts efficiently in a continual learning setting remains an open challenge for current Artificial Intelligence algorithms as persistent model retraining is necessary. Inspired by the Parallel Distributed Processing learning and the Complementary Learning Systems theories, we develop a computational model that is able to expand its previously learned concepts efficiently to new domains using a few labeled samples. We couple the new form of a concept to its past learned forms in an embedding space for effective continual learning. Doing so, a generative distribution is learned such that it is shared across the tasks in the embedding space and models the abstract concepts. This procedure enables the model to generate pseudo-data points to replay the past experience to tackle catastrophic forgetting.

Rapid online adaptation to changing tasks is an important problem in machine learning and, recently, a focus of meta-reinforcement learning. However, reinforcement learning (RL) algorithms struggle in POMDP environments because the state of the system, essential in a RL framework, is not always visible. Additionally, hand-designed meta-RL architectures may not include suitable computational structures for specific learning problems. The evolution of online learning mechanisms, on the contrary, has the ability to incorporate learning strategies into an agent that can (i) evolve memory when required and (ii) optimize adaptation speed to specific online learning problems. In this paper, we exploit the highly adaptive nature of neuromodulated neural networks to evolve a controller that uses the latent space of an autoencoder in a POMDP. The analysis of the evolved networks reveals the ability of the proposed algorithm to acquire inborn knowledge in a variety of aspects such as the detection of cues that reveal implicit rewards, and the ability to evolve location neurons that help with navigation. The integration of inborn knowledge and online plasticity enabled fast adaptation and better performance in comparison to some non-evolutionary meta-reinforcement learning algorithms. The algorithm proved also to succeed in the 3D gaming environment Malmo Minecraft.

Despite huge success, deep networks are unable to learn effectively in sequential multitask learning settings as they forget the past learned tasks after learning new tasks. Inspired from complementary learning systems theory, we address this challenge by learning a generative model that couples the current task to the past learned tasks through a discriminative embedding space. We learn an abstract generative distribution in the embedding that allows generation of data points to represent past experience. We sample from this distribution and utilize experience replay to avoid forgetting and simultaneously accumulate new knowledge to the abstract distribution in order to couple the current task with past experience. We demonstrate theoretically and empirically that our framework learns a distribution in the embedding, which is shared across all tasks, and as a result tackles catastrophic forgetting.

We provide here a novel method, called hypercolumn sparsification, to achieve high recognition performance for convolutional neural networks (CNNs) despite low-precision weights and activities during both training and test phases. This method is applicable to any CNN architecture that operates on signal patterns (e.g., audio, image, video) to extract information such as class membership. It operates on the stack of feature maps in each of the cascading feature matching and pooling layers through the processing hierarchy of the CNN by an explicit competitive process ( k -WTA, winner take all) that generates a sparse feature vector at each spatial location. This principle is inspired by local brain circuits, where neurons tuned to respond to different patterns in the incoming signals from an upstream region inhibit each other using interneurons, such that only the ones that are maximally activated survive the quenching threshold. We show this process of sparsification is critical for probabilistic learning of low-precision weights and bias terms, thereby making pattern recognition amenable for energy-efficient hardware implementations. Further, we show that hypercolumn sparsification could lead to more data-efficient learning as well as having an emergent property of significantly pruning down the number of connections in the network. A theoretical account and empirical analysis are provided to understand these effects better.

Place cells in the hippocampus of higher mammals are critical for spatial navigation. Recent modeling clarifies how this may be achieved by how grid cells in the medial entorhinal cortex (MEC) input to place cells. Grid cells exhibit hexagonal grid firing patterns across space in multiple spatial scales along the MEC dorsoventral axis. Signals from grid cells of multiple scales combine adaptively to activate place cells that represent much larger spaces than grid cells. But how do grid cells learn to fire at multiple positions that form a hexagonal grid, and with spatial scales that increase along the dorsoventral axis? In vitro recordings of medial entorhinal layer II stellate cells have revealed subthreshold membrane potential oscillations (MPOs) whose temporal periods, and time constants of excitatory postsynaptic potentials (EPSPs), both increase along this axis. Slower (faster) subthreshold MPOs and slower (faster) EPSPs correlate with larger (smaller) grid spacings and field widths. A self-organizing map neural model explains how the anatomical gradient of grid spatial scales can be learned by cells that respond more slowly along the gradient to their inputs from stripe cells of multiple scales, which perform linear velocity path integration. The model cells also exhibit MPO frequencies that covary with their response rates. The gradient in intrinsic rhythmicity is thus not compelling evidence for oscillatory interference as a mechanism of grid cell firing. A response rate gradient combined with input stripe cells that have normalized receptive fields can reproduce all known spatial and temporal properties of grid cells along the MEC dorsoventral axis. This spatial gradient mechanism is homologous to a gradient mechanism for temporal learning in the lateral entorhinal cortex and its hippocampal projections. Spatial and temporal representations may hereby arise from homologous mechanisms, thereby embodying a mechanistic "neural relativity" that may clarify how episodic memories are learned.