In this work we present Deep Reinforcement Learning (DRL) training of directional locomotion for low-cost quadrupedal robots in the real world. In particular, we exploit randomization of heading that the robot must follow to foster exploration of action-state transitions most useful for learning both forward locomotion as well as course adjustments. Changing the heading in episode resets to current yaw plus a random value drawn from a normal distribution yields policies able to follow complex trajectories involving frequent turns in both directions as well as long straight-line stretches. By repeatedly changing the heading, this method keeps the robot moving within the training platform and thus reduces human involvement and need for manual resets during the training. Real world experiments on a custom-built, low-cost quadruped demonstrate the efficacy of our method with the robot successfully navigating all validation tests. When trained with other approaches, the robot only succeeds in forward locomotion test and fails when turning is required.

Object reorientation is a critical task for robotic grippers, especially when manipulating objects within constrained environments. The task poses significant challenges for motion planning due to the high-dimensional output actions with the complex input information, including unknown object properties and nonlinear contact forces. Traditional approaches simplify the problem by reducing degrees of freedom, limiting contact forms, or acquiring environment/object information in advance, which significantly compromises adaptability. To address these challenges, we deconstruct the complex output actions into three fundamental types based on tactile sensing: task-oriented actions, constraint-oriented actions, and coordinating actions. These actions are then optimized online using gradient optimization to enhance adaptability. Key contributions include simplifying contact state perception, decomposing complex gripper actions, and enabling online action optimization for handling unknown objects or environmental constraints. Experimental results demonstrate that the proposed method is effective across a range of everyday objects, regardless of environmental contact. Additionally, the method exhibits robust performance even in the presence of unknown contacts and nonlinear external disturbances.

Constrained pathfinding is a well-studied, yet challenging network optimisation problem that can be seen in a broad range of real-world applications. Pathfinding with multiple resource limits, which is known as the Resource Constrained Shortest Path Problem (RCSP), aims to plan a cost-optimum path subject to limited usage of resources. Given the recent advances in constrained and multi-criteria search with A*, this paper introduces a new resource constrained search framework on the basis of A* to tackle RCSP in large networks, even in the presence of negative cost and negative resources. We empirically evaluate our new algorithm on a set of large instances and show up to two orders of magnitude faster performance compared to state-of-the-art RCSP algorithms in the literature.

Spectral bias is a significant phenomenon in neural network training and can be explained by neural tangent kernel (NTK) theory. In this work, we develop the NTK theory for deep neural networks with physics-informed loss, providing insights into the convergence of NTK during initialization and training, and revealing its explicit structure. We find that, in most cases, the differential operators in the loss function do not induce a faster eigenvalue decay rate and stronger spectral bias. Some experimental results are also presented to verify the theory.

Domain randomization has emerged as a fundamental technique in reinforcement learning (RL) to facilitate the transfer of policies from simulation to real-world robotic applications. Many existing domain randomization approaches have been proposed to improve robustness and sim2real transfer. These approaches rely on wide randomization ranges to compensate for the unknown actual system parameters, leading to robust but inefficient real-world policies. In addition, the policies pretrained in the domain-randomized simulation are fixed after deployment due to the inherent instability of the optimization processes based on RL and the necessity of sampling exploitative but potentially unsafe actions on the real system. This limits the adaptability of the deployed policy to the inevitably changing system parameters or environment dynamics over time. We leverage safe RL and continual learning under domain-randomized simulation to address these limitations and enable safe deployment-time policy adaptation in real-world robot control. The experiments show that our method enables the policy to adapt and fit to the current domain distribution and environment dynamics of the real system while minimizing safety risks and avoiding issues like catastrophic forgetting of the general policy found in randomized simulation during the pretraining phase. Videos and supplementary material are available at https://safe-cda.github.io/.

While Large Language Models (LLMs) have significantly advanced natural language processing, aligning them with human preferences remains an open challenge. Although current alignment methods rely primarily on explicit feedback, eye-tracking (ET) data offers insights into real-time cognitive processing during reading. In this paper, we present OASST-ETC, a novel eye-tracking corpus capturing reading patterns from 24 participants, while evaluating LLM-generated responses from the OASST1 dataset. Our analysis reveals distinct reading patterns between preferred and non-preferred responses, which we compare with synthetic eye-tracking data. Furthermore, we examine the correlation between human reading measures and attention patterns from various transformer-based models, discovering stronger correlations in preferred responses. This work introduces a unique resource for studying human cognitive processing in LLM evaluation and suggests promising directions for incorporating eye-tracking data into alignment methods. The dataset and analysis code are publicly available.

When employing an evolutionary algorithm to optimize a neural networks architecture, developers face the added challenge of tuning the evolutionary algorithm's own hyperparameters - population size, mutation rate, cloning rate, and number of generations. This paper introduces Neuvo Ecological Neural Architecture Search (ENAS), a novel method that incorporates these evolutionary parameters directly into the candidate solutions' phenotypes, allowing them to evolve dynamically alongside architecture specifications. Experimental results across four binary classification datasets demonstrate that ENAS not only eliminates manual tuning of evolutionary parameters but also outperforms competitor NAS methodologies in convergence speed (reducing computational time by 18.3%) and accuracy (improving classification performance in 3 out of 4 datasets). By enabling "greedy individuals" to optimize resource allocation based on fitness, ENAS provides an efficient, self-regulating approach to neural architecture search.

Mechanistic interpretability has made great strides in identifying neural network features (e.g., directions in hidden activation space) that mediate concepts(e.g., the birth year of a person) and enable predictable manipulation. Distributed alignment search (DAS) leverages supervision from counterfactual data to learn concept features within hidden states, but DAS assumes we can afford to conduct a brute force search over potential feature locations. To address this, we present HyperDAS, a transformer-based hypernetwork architecture that (1) automatically locates the token-positions of the residual stream that a concept is realized in and (2) constructs features of those residual stream vectors for the concept. In experiments with Llama3-8B, HyperDAS achieves state-of-the-art performance on the RAVEL benchmark for disentangling concepts in hidden states. In addition, we review the design decisions we made to mitigate the concern that HyperDAS (like all powerful interpretabilty methods) might inject new information into the target model rather than faithfully interpreting it.

In the context of pressing climate change challenges and the significant biodiversity loss among arthropods, automated taxonomic classification from organismal images is a subject of intense research. However, traditional AI pipelines based on deep neural visual architectures such as CNNs or ViTs face limitations such as degraded performance on the long-tail of classes and the inability to reason about their predictions. We integrate image captioning and retrieval-augmented generation (RAG) with large language models (LLMs) to enhance biodiversity monitoring, showing particular promise for characterizing rare and unknown arthropod species. While a naive Vision-Language Model (VLM) excels in classifying images of common species, the RAG model enables classification of rarer taxa by matching explicit textual descriptions of taxonomic features to contextual biodiversity text data from external sources. The RAG model shows promise in reducing overconfidence and enhancing accuracy relative to naive LLMs, suggesting its viability in capturing the nuances of taxonomic hierarchy, particularly at the challenging family and genus levels. Our findings highlight the potential for modern vision-language AI pipelines to support biodiversity conservation initiatives, emphasizing the role of comprehensive data curation and collaboration with citizen science platforms to improve species identification, unknown species characterization and ultimately inform conservation strategies.

The choice of neural network features can have a large impact on both the accuracy and speed of the network. Despite the current industry shift towards large transformer models, specialized binary classifiers remain critical for numerous practical applications where computational efficiency and low latency are essential. Neural network features tend to be developed homogeneously, resulting in slower or less accurate networks when testing against multiple datasets. In this paper, we show the effectiveness of Neuvo NAS+ a novel Python implementation of an extended Neural Architecture Search (NAS+) which allows the user to optimise the training parameters of a network as well as the network's architecture. We provide an in-depth analysis of the importance of catering a network's architecture to each dataset. We also describe the design of the Neuvo NAS+ system that selects network features on a task-specific basis including network training hyper-parameters such as the number of epochs and batch size. Results show that the Neuvo NAS+ task-specific approach significantly outperforms several machine learning approaches such as Naive Bayes, C4.5, Support Vector Machine and a standard Artificial Neural Network for solving a range of binary classification problems in terms of accuracy. Our experiments demonstrate substantial diversity in evolved network architectures across different datasets, confirming the value of task-specific optimization. Additionally, Neuvo NAS+ outperforms other evolutionary algorithm optimisers in terms of both accuracy and computational efficiency, showing that properly optimized binary classifiers can match or exceed the performance of more complex models while requiring significantly fewer computational resources.