Humans use signs, e.g., sentences in a spoken language, for communication and thought. Hence, symbol systems like language are crucial our with other agents adaptation to real-world environment. The we human society adaptively dynamically change over time. In the context of artificial intelligence (AI) cognitive systems, grounding problem has been regarded as one central problems related {\it symbols}. However, was originally posed connect symbolic AI sensorimotor information did not...

Although a manipulator must interact with objects in terms of their full complexity, it is the qualitative structure an environment and relationships between them which define composition that environment, allow for construction efficient plans to enable completion various elaborate tasks. In this paper we present algorithm redescribes scene layered representation, from labeled point clouds scene. The representation includes description objects, as well symbolic them. This achieved by...

The high variability of fingerprint data (owing to, e.g., differences in quality, moisture conditions, and scanners) makes the task minutiae extraction challenging, particularly when approached from a stance that relies on tunable algorithmic components, such as image enhancement. We pose machine learning problem propose deep neural network - MENet, for Minutiae Extraction Network to learn data-driven representation points. By using existing capabilities several algorithms, we establish...

A key challenge in many reinforcement learning problems is delayed rewards, which can significantly slow down learning. Although reward shaping has previously been introduced to accelerate by bootstrapping an agent with additional information, this lead convergence. We present a novel Bayesian framework that augments the distribution prior beliefs decay experience. Formally, we prove under suitable conditions Markov decision process augmented our consistent optimal policy of original MDP...

We present a method for segmenting set of unstructured demonstration trajectories to discover reusable skills using inverse reinforcement learning (IRL). Each skill is characterised by latent reward function which the demonstrator assumed be optimizing. The boundaries and number making up each are unknown. use Bayesian nonparametric approach propose segmentations maximum entropy infer functions from segments. This produces Markov Decision Processes (MDPs) that best describe input...

Diverse freshwater biological communities are threatened by invasive aquatic alien plant (IAAP) invasions and consequently, cost countries millions to manage. The effective management of these IAAP necessitates their frequent reliable monitoring across a broad extent over long-term. Here, we introduce apply approach that meet criteria is based on three-stage hierarchical classification firstly detect water, then vegetation finally water hyacinth (Pontederia crassipes, previously Eichhornia...

Monte Carlo Tree Search (MCTS) is a family of directed search algorithms that has gained widespread attention in recent years. Despite the vast amount research into MCTS, effect modifications on algorithm, as well manner which it performs various domains, still not yet fully known. In particular, using knowledge-heavy rollouts MCTS remains poorly understood, with surprising results demonstrating better-informed often result worse-performing agents. We present experimental evidence suggesting...

We propose an architecture as a robot»s decision-making mechanism to anticipate human»s state of mind, and so plan accordingly during human-robot collaboration task. At the core lies novel stochastic that implements partially observable Markov decision process anticipating mind in two-stages. In first stage it anticipates task related availability, intent (motivation), capability collaboration. second, further reasons about these states true need for help. Our contribution ability our model...

Personalised education is one of the domains that can greatly benefit from most recent advances in Artificial Intelligence (AI) and Large Language Models (LLM). However, it also challenging applications due to cognitive complexity teaching effectively while personalising learning experience suit independent learners. We hypothesise promising approach excelling such demanding use cases using a \emph{society minds}. In this chapter, we present TrueReason, an exemplar personalised system...

An important property of long-lived agents is the ability to reuse existing knowledge solve new tasks. appealing approach towards obtaining such by leveraging logical composition over tasks, where tasks are defined applying logic operators previously-solved ones. This particularly powerful since it provides a human-understandable mechanism for task specification. However, no unifying formalism and generalising combinatorially them has yet been developed. We address problem formally defining...

A limitation for collaborative robots (cobots) is their lack of ability to adapt human partners, who typically exhibit an immense diversity behaviors. We present autonomous framework as a cobot's real-time decision-making mechanism anticipate variety characteristics and behaviors, including errors, toward personalized collaboration. Our handles such behaviors in two levels: 1) short-term are adapted through our novel Anticipatory Partially Observable Markov Decision Process (A-POMDP) models,...

Combining reinforcement learning with language grounding is challenging as the agent needs to explore environment while simultaneously multiple language-conditioned tasks. To address this, we introduce a novel method: compositionally-enabled (CERLLA). Our method reduces sample complexity of tasks specified by leveraging compositional policy representations and semantic parser trained using in-context learning. We evaluate our approach in an requiring function approximation demonstrate...

Learning of robot kinematic and dynamic models from data has attracted much interest recently as an alternative to manually defined models. However, the amount required learn these becomes large when number degrees freedom increases collecting it can be a time-intensive process. We employ transfer learning techniques in order speed up models, by using additional obtained other robots. propose method for approximating non-linear mappings between manifolds, which we call Local Procrustes...

The computational complexity of learning in sequential decision problems grows exponentially with the number actions available to agent at each state. We present a method for accelerating this process by action priors that express usefulness These are learned from set different optimal policies many tasks same state space, and used bias exploration away less useful actions. This is shown improve performance domain but goals. extend our base on perceptual cues rather than absolute states,...

In this paper we use content-based features to perform automatic classification of music pieces into genres. We categorise these four groups: extracted from the Fourier transform's magnitude spectrum, designed inform on tempo, pitch-related features, and chordal features. a novel thorough exploration performance for different feature representations, including mean standard deviation its distribution, by histogram various bin sizes, using mel-frequency cepstral coefficients. Finally, uses...

Training sparse networks to converge the same performance as dense neural architectures has proven be elusive. Recent work suggests that initialization is key. However, while this direction of research had some success, focusing on alone appears inadequate. In paper, we take a broader view training and consider role regularization, optimization, architecture choices models. We propose simple experimental framework, Same Capacity Sparse vs Dense Comparison (SC-SDC), allows for fair comparison...

We present algorithms to effectively represent a set of Markov decision processes (MDPs), whose optimal policies have already been learned, by smaller source subset for lifelong, policy-reuse-based transfer learning in reinforcement learning. This is necessary when the number previous tasks large and cost measuring similarity counteracts benefit transfer. The forms an `$ε$-net' over original MDPs, sense that each MDP $M_p$, there $M^s$ policy has $

Natural Language Understanding (NLU) is considered a core component in implementing dialogue systems. NLU has been greatly enhanced by deep learning techniques such as word embeddings and neural network architectures, but current NLP methods for Arabic language action classification or semantic decoding mostly based on handcrafted rule-based systems that use feature engineering, without the benefit of any form distributed representation words. This paper presents an approach to text Named...