This paper introduces a novel system for in-home cognitive health assessment using ambient sensors and a machine learning technology that can robustly detect mild cognitive impairment (MCI) despite limited available data. The learned model can explain the aspects of individuals' daily lives led to the prediction, while reliably predicting MCI, providing more insights to healthcare workers for further clinical interventions. We developed the robust transparent machine learning model, based on fusion adaptive resonance theory (Fusion ART) neural network to learn individuals' daily patterns of activity from continuous sensor data in terms of a suite of digital biomarkers reflecting four key domains: physical, daily activity, cognitive engagement, and sleep patterns. Based on a longitudinal study of over one hundred participants, deployed with non-intrusive sensors in their homes to undergo parallel clinical evaluation across a period of five years, our model successfully identified individuals with MCI, achieving high predictive accuracy regardless the noisy and sparse availability of data. As a transparent neural network, the learned model can also be interpreted as classification rules to distinguish MCI from normal cognition (NC) cases based on the digital biomarkers. These results demonstrate that passively collected, sensor-derived digital biomarkers can be leveraged to indicate cognitive status and potentially providing clinically meaningful insights on the impairment conditions. We also discuss the practical challenges and lessons learned from this real-world deployment to inform future large-scale implementations of such AI-driven health monitoring systems.

Current large language models (LLMs) exhibit significant deficiencies in episodic memory tasks including encoding, storing, and retrieving specific information from temporally dependent events over a long period of time. Recent approaches to handle memory tasks in LLMs, such as in-context learning, retrieval-augmented generation (RAG), and fine-tuning, may resolve the long-term retention issues, but are still inadequate to handle tasks requiring chronological awareness of the stored information. We introduce Agentic Retrieval with Temporal-Episodic Memory (ARTEM), a hybrid LLM-based agent architecture integrating LLMs with a self-organizing neural network named Spatial-Temporal Episodic Memory (STEM), designed to handle episodic memory tasks. Our approach employs LLMs for event extraction from the inputs to represent temporal, spatial, entitative, and semantic information that may facilitate future retrieval, aside from generating outputs or direct responses. The extracted events can then be encoded vectorially and stored in a fast and stable manner in the episodic memory through an instance-based incremental learning in STEM. STEM supports precise episodes retrieval and helps reduce computational overhead in generating the appropriate responses by LLMs. Evaluation on standardized episodic memory benchmarks across four tasks—partial cue retrieval, epistemic uncertainty detection, recent event identification, and chronological recall—demonstrates superior performance of ARTEM compared to in-context learning, RAG, and fine-tuning in various popular LLMs.

Multi-agent reinforcement learning (MARL) has demonstrated remarkable success in collaborative tasks, yet faces significant challenges in scaling to complex scenarios requiring sustained planning and coordination across long horizons. While hierarchical approaches help decompose these tasks, they typically rely on hand-crafted subtasks and domain-specific knowledge, limiting their generalizability. We present L2M2, a novel hierarchical framework that leverages large language models (LLMs) for high-level strategic planning and MARL for low-level execution. L2M2 enables zero-shot planning that supports both end-to-end training and direct integration with pre-trained MARL models. Experiments in the VMAS environment demonstrate that L2M2's LLM-guided MARL achieves superior performance while requiring less than 20% of the training samples compared to baseline methods. In the MOSMAC environment, L2M2 demonstrates strong performance with pre-defined subgoals and maintains substantial effectiveness without subgoals - scenarios where baseline methods consistently fail. Analysis through kernel density estimation reveals L2M2's ability to automatically generate appropriate navigation plans, demonstrating its potential for addressing complex multi-agent coordination tasks.

One main challenge in constructing a Knowledge Graph (KG) is to deal with ambiguity. Specifically, an entity in the graph can be assigned with multiple meanings while two or more entities considered to have different meanings may actually be the same. Assigning an entity with the correct meaning may involve re-evaluation of its relevant contexts. This costly operation typically involves searching for other similar entities within the KG such that the context can be determined. In this article, a new model called DisambiguART is proposed leveraging multi-channel matching and inference in a self-organizing neural network for sense disambiguation in KGs. Unlike other disambiguation methods that rely on representation learning to identify the relevant contexts whereby similarities among entities are learned, DisambiguART extends the working principle of multi-channel Adaptive Resonance Theory (ART) to conduct inferences directly over the graph representation through bi-directional interactions of bottom-up activations and top-down matching to find similar entities and select the correct meaning according to the right context. The proposed method is evaluated on the tasks of entity sense disambiguation in three domain KGs (jet engine, biomedical, and kinship) and author name disambiguation in bibliographic KGs, demonstrating the effectiveness and efficiency of DisambiguART against the state-of-the-art methods.

Learning of Activities of Daily Living (ADLs) provides insights into an individual’s habits, lifestyle, and well-being. However, it is crucial to address data privacy concerns in practical situations when learning the ADL routines of individuals. In this paper, we introduce FedSTEM-ADL, a federated spatial-temporal episodic memory model to address this privacy issue. FedSTEM-ADL utilizes a federation of Spatial-Temporal Episodic Memory for ADLs (STEM-ADL) for federated learning, wherein multiple local STEM-ADL models from individual users are combined into a global model while preserving the privacy of the original data. Specifically, each local model is designed to learn the spatio-temporal ADL routines of an individual user, representing them as ADL events and sequences of such events as episode patterns. The global model then integrates the local models without referring to the underlying individual data, thus addressing privacy concerns in multi-user ADL analysis. We conduct a series of experiments based on both pseudo and real-world multi-user ADL datasets. The results show that FedSTEM-ADL is able to learn global ADL models in an efficient manner and consistently outperforms the baseline models in the task of next ADL event prediction.

Learning to reach long-horizon goals in spatial traversal tasks is a significant challenge for autonomous agents. Recent subgoal graph-based planning methods address this challenge by decomposing a goal into a sequence of shorter-horizon subgoals. These methods, however, use arbitrary heuristics for sampling or discovering subgoals, which may not conform to the cumulative reward distribution. Moreover, they are prone to learning erroneous connections (edges) between subgoals, especially those lying across obstacles. To address these issues, this article proposes a novel subgoal graph-based planning method called learning subgoal graph using value-based subgoal discovery and automatic pruning (LSGVP). The proposed method uses a subgoal discovery heuristic that is based on a cumulative reward (value) measure and yields sparse subgoals, including those lying on the higher cumulative reward paths. Moreover, LSGVP guides the agent to automatically prune the learned subgoal graph to remove the erroneous edges. The combination of these novel features helps the LSGVP agent to achieve higher cumulative positive rewards than other subgoal sampling or discovery heuristics, as well as higher goal-reaching success rates than other state-of-the-art subgoal graph-based planning methods.

Spatial mapping and navigation are critical cognitive functions of autonomous agents, enabling one to learn an internal representation of an environment and move through space with real-time sensory inputs, such as visual observations. Existing models for vision-based mapping and navigation, however, suffer from memory requirements that increase linearly with exploration duration and indirect path following behaviors. This article presents e -TM, a self-organizing neural network-based framework for incremental topological mapping and navigation. e -TM models the exploration trajectories explicitly as episodic memory, wherein salient landmarks are sequentially extracted as "events" from streaming observations. A memory consolidation procedure then performs a playback mechanism and transfers the embedded knowledge of the environmental layout into spatial memory, encoding topological relations between landmarks. Fusion adaptive resonance theory (ART) networks, as the building block of the two memory modules, can generalize multiple input patterns into memory templates and, therefore, provide a compact spatial representation and support the discovery of novel shortcuts through inferences. For navigation, e -TM applies a transfer learning paradigm to integrate human demonstrations into a pretrained locomotion network for smoother movements. Experimental results based on VizDoom, a simulated 3-D environment, have shown that, compared to semiparametric topological memory (SPTM), a state-of-the-art model, e -TM reduces the time costs of navigation significantly while learning much sparser topological graphs.

Hierarchical reinforcement learning (HRL) is a promising approach to perform long-horizon goal-reaching tasks by decomposing the goals into subgoals. In a holistic HRL paradigm, an agent must autonomously discover such subgoals and also learn a hierarchy of policies that uses them to reach the goals. Recently introduced end-to-end HRL methods accomplish this by using the higher-level policy in the hierarchy to directly search the useful subgoals in a continuous subgoal space. However, learning such a policy may be challenging when the subgoal space is large. We propose integrated discovery of salient subgoals (LIDOSS), an end-to-end HRL method with an integrated subgoal discovery heuristic that reduces the search space of the higher-level policy, by explicitly focusing on the subgoals that have a greater probability of occurrence on various state-transition trajectories leading to the goal. We evaluate LIDOSS on a set of continuous control tasks in the MuJoCo domain against hierarchical actor critic (HAC), a state-of-the-art end-to-end HRL method. The results show that LIDOSS attains better goal achievement rates than HAC in most of the tasks.

Goal recognition for path planning is an important task of intention identification and situation awareness, requiring an observer to predict the goal of an evader given observations of its movements. While existing models based on planning or Markov Decision Process (MDP) show superior performance over traditional library based methods, they require much effort in model design and can hardly provide legible decision rules for their users. To make the system more user-friendly while preserving accuracy of goal inference, this paper proposes a novel self-organizing neural network based inference model, which learns compact rule sets through generalizing the streaming observations of an evader. More critically, the system manifests a high level of interpretability with the linguistic if-then rule base, making it easily comprehensible for human decision makers. We conducted extensive experiments on a large-scale real-world road network. Results show that the proposed model produces accuracy comparable to those of two state-of-the-art methods while uniquely providing legible inference rules and strong robustness against multiple goals with missing data.

Most of today's AI technologies are geared towards mastering specific tasks performance through learning from a huge volume of data. However, less attention has still been given to make the AI understand its own purposes or be responsible socially. In this paper, a new model of agent is presented with the capacity to represent itself as a distinct individual with identity, a mind of its own, unique experiences, and social lives. In this way, the agent can interact with its surroundings and other agents seamlessly and meaningfully. A practical framework for developing an agent architecture with this model of self and self-awareness is proposed allowing self to be ascribed to an existing intelligent agent architecture in general to enable its social ability, interactivity, and co-presence with others. Possible applications are discussed with some exemplifying cases based on an implementation of a conversational agent.

Self-awareness is a crucial feature for a sociable agent or robot to better interact with humans. In a futuristic scenario, a conversational agent may occasionally be asked for its own opinion or suggestion based on its own thought, feelings, or experiences as if it is an individual with identity, personality, and social life. In moving towards that direction, in this paper, a brain inspired model of self-awareness is presented that allows an agent to learn to attend to different aspects of self as an individual with identity, physical embodiment, mental states, experiences, and reflections on how others may think about oneself. The model is built and realized on a NAO humanoid robotic platform to investigate the role of this capacity of self-awareness on the robot's learning and interactivity.

Developing a complex intelligent system by abstracting their behaviors, functionalities, and reasoning mechanisms can be tedious and time consuming. In this paper, we present a framework for developing an application or software system based on smart autonomous components that collaborate with the developer or user to realize the entire system. Inspired by teachable approaches and programming-by-demonstration methods in robotics and end-user development, we treat intelligent agents as teachable components that make up the system to be built. Each agent serves different functionalities and may have prebuilt operations to accomplish its own design objectives. However, each agent may also be equipped with in-built social-cognitive traits to interact with the user or other agents in order to adapt its own operations, objectives, and relationships with others. The results of adaptation can be in the form of groups or multiagent systems as new aggregated components. This approach is made to tackle the difficulties in completely programming the entire system by allowing the user to teach the components toward the desired behaviors in the situated context of the application. We exemplify this novel method with cases in the domains of human-like agents in virtual environment and agents for in-house caregiving.

This paper presents a self-organizing approach to the learning of procedural and declarative knowledge in parallel using independent but interconnected memory models. The proposed system, employing fusion Adaptive Resonance Theory (fusion ART) network as a building block, consists of a declarative memory module, that learns both episodic traces and semantic knowledge in real time, as well as a procedural memory module that learns reactive responses to its environment through reinforcement learning. More importantly, the proposed multi-memory system demonstrates how the various memory modules transfer knowledge and cooperate with each other for a higher overall performance. We present experimental studies, wherein the proposed system is tasked to learn the procedural and declarative knowledge for an autonomous agent playing in a first person game environment called Unreal Tournament. Our experimental results show that the multi-memory system is able to enhance the performance of the agent in a real time environment by utilizing both its procedural and declarative knowledge.

This paper presents a neural model that learns episodic traces in response to a continuous stream of sensory input and feedback received from the environment. The proposed model, based on fusion adaptive resonance theory (ART) network, extracts key events and encodes spatio-temporal relations between events by creating cognitive nodes dynamically. The model further incorporates a novel memory search procedure, which performs a continuous parallel search of stored episodic traces. Combined with a mechanism of gradual forgetting, the model is able to achieve a high level of memory performance and robustness, while controlling memory consumption over time. We present experimental studies, where the proposed episodic memory model is evaluated based on the memory consumption for encoding events and episodes as well as recall accuracy using partial and erroneous cues. Our experimental results show that: 1) the model produces highly robust performance in encoding and recalling events and episodes even with incomplete and noisy cues; 2) the model provides enhanced performance in a noisy environment due to the process of forgetting; and 3) compared with prior models of spatio-temporal memory, our model shows a higher tolerance toward noise and errors in the retrieval cues.

This paper presents a neural model that learns episodic traces in response to a continual stream of sensory input and feedback received from the environment. The proposed model, based on fusion Adaptive Resonance Theory (fusion ART) network, extracts key events and encodes spatio-temporal relations between events by creating cognitive nodes dynamically. The model further incorporates a novel memory search procedure, which performs parallel search of stored episodic traces continuously. Comparing with prior systems, the proposed episodic memory model presents a robust approach to encoding key events and episodes and recalling them using partial and erroneous cues. We present experimental studies, wherein the model is used to learn episodic memory of an agent's experience in a first person game environment called Unreal Tournament. Our experimental results show that the model produces highly robust performance in encoding and recalling events and episodes even with incomplete and noisy cues.