Large language models (LLMs) have recently revolutionized automated text understanding and generation. The performance of these models relies on the high number of parameters of the underlying neural architectures, which allows LLMs to memorize part of the vast quantity of data seen during the training. This paper investigates whether and to what extent general-purpose pre-trained LLMs have memorized information from known ontologies. Our results show that LLMs partially know ontologies: they can, and do indeed, memorize concepts from ontologies mentioned in the text, but the level of memorization of their concepts seems to vary proportionally to their popularity on the Web, the primary source of their training material. We additionally propose new metrics to estimate the degree of memorization of ontological information in LLMs by measuring the consistency of the output produced across different prompt repetitions, query languages, and degrees of determinism.
The salient information of an infrared image and the abundant texture of a visible image can be fused to obtain a comprehensive image. As can be known, the current fusion methods based on Transformer techniques for infrared and visible (IV) images have exhibited promising performance. However, the attention mechanism of the previous Transformer-based methods was prone to extract common information from source images without considering the discrepancy information, which limited fusion performance. In this paper, by reevaluating the cross-attention mechanism, we propose an alternate Transformer fusion network (ATFuse) to fuse IV images. Our ATFuse consists of one discrepancy information injection module (DIIM) and two alternate common information injection modules (ACIIM). The DIIM is designed by modifying the vanilla cross-attention mechanism, which can promote the extraction of the discrepancy information of the source images. Meanwhile, the ACIIM is devised by alternately using the vanilla cross-attention mechanism, which can fully mine common information and integrate long dependencies. Moreover, the successful training of ATFuse is facilitated by a proposed segmented pixel loss function, which provides a good trade-off for texture detail and salient structure preservation. The qualitative and quantitative results on public datasets indicate our ATFFuse is effective and superior compared to other state-of-the-art methods.
The self-attention mechanism utilizes large implicit weight matrices, programmed through dot product-based activations with very few trainable parameters, to enable long sequence modeling. In this paper, we investigate the possibility of discarding residual learning by employing large implicit kernels to achieve full context interaction at each layer of the network. To accomplish it, we introduce coordinate-based implicit MLPs as a slow network to generate hyper-kernels for another fast convolutional network. To get context-varying weights for fast dynamic encoding, we propose a $\mathrm{Hyper}\mathcal{Z{\cdot}Z{\cdot}W}$ operator that connects hyper-kernels ($\mathcal{W}$) and hidden activations ($\mathcal{Z}$) through simple elementwise multiplication, followed by convolution of $\mathcal{Z}$ using the context-dependent $\mathcal{W}$. Based on this design, we present a novel Terminator architecture that integrates hyper-kernels of different sizes to produce multi-branch hidden representations for enhancing the feature extraction capability of each layer. Additionally, a bottleneck layer is employed to compress the concatenated channels, allowing only valuable information to propagate to the subsequent layers. Notably, our model incorporates several innovative components and exhibits excellent properties, such as introducing local feedback error for updating the slow network, stable zero-mean features, faster training convergence, and fewer model parameters. Extensive experimental results on pixel-level 1D and 2D image classification benchmarks demonstrate the superior performance of our architecture.
As spiking neural networks receive more attention, we look toward applications of this computing paradigm in fields other than computer vision and signal processing. One major field, underexplored in the neuromorphic setting, is Natural Language Processing (NLP), where most state-of-the-art solutions still heavily rely on resource-consuming and power-hungry traditional deep learning architectures. Therefore, it is compelling to design NLP models for neuromorphic architectures due to their low energy requirements, with the additional benefit of a more human-brain-like operating model for processing information. However, one of the biggest issues with bringing NLP to the neuromorphic setting is in properly encoding text into a spike train so that it can be seamlessly handled by both current and future SNN architectures. In this paper, we compare various methods of encoding text as spikes and assess each method's performance in an associated SNN on a downstream NLP task, namely, sentiment analysis. Furthermore, we go on to propose a new method of encoding text as spikes that outperforms a widely-used rate-coding technique, Poisson rate-coding, by around 13\% on our benchmark NLP tasks. Subsequently, we demonstrate the energy efficiency of SNNs implemented in hardware for the sentiment analysis task compared to traditional deep neural networks, observing an energy efficiency increase of more than 32x during inference and 60x during training while incurring the expected energy-performance tradeoff.
This paper, the fourth part of a series of papers on the arithmetic average (AA) density fusion approach and its application for target tracking, addresses the intricate challenge of distributed heterogeneous multisensor multitarget tracking, where each inter-connected sensor operates a probability hypothesis density (PHD) filter, a multiple Bernoulli (MB) filter or a labeled MB (LMB) filter and they cooperate with each other via information fusion. Earlier papers in this series have proven that the proper AA fusion of these filters is all exactly built on averaging their respective unlabeled/labeled PHDs. Based on this finding, two PHD-AA fusion approaches are proposed via variational minimization of the upper bound of the Kullback-Leibler divergence between the local and multi-filter averaged PHDs subject to cardinality consensus based on the Gaussian mixture implementation, enabling heterogeneous filter cooperation. One focuses solely on fitting the weights of the local Gaussian components (L-GCs), while the other simultaneously fits all the parameters of the L-GCs at each sensor, both seeking average consensus on the unlabeled PHD, irrespective of the specific posterior form of the local filters. For the distributed peer-to-peer communication, both the classic consensus and flooding paradigms have been investigated. Simulations have demonstrated the effectiveness and flexibility of the proposed approaches in both homogeneous and heterogeneous scenarios.
Conversational agents are increasingly used to address emotional needs on top of information needs. One use case of increasing interest are counselling-style mental health and behaviour change interventions, with large language model (LLM)-based approaches becoming more popular. Research in this context so far has been largely system-focused, foregoing the aspect of user behaviour and the impact this can have on LLM-generated texts. To address this issue, we share a dataset containing text-based user interactions related to behaviour change with two GPT-4-based conversational agents collected in a preregistered user study. This dataset includes conversation data, user language analysis, perception measures, and user feedback for LLM-generated turns, and can offer valuable insights to inform the design of such systems based on real interactions.
This paper introduces LeTO, a method for learning constrained visuomotor policy via differentiable trajectory optimization. Our approach uniquely integrates a differentiable optimization layer into the neural network. By formulating the optimization layer as a trajectory optimization problem, we enable the model to end-to-end generate actions in a safe and controlled fashion without extra modules. Our method allows for the introduction of constraints information during the training process, thereby balancing the training objectives of satisfying constraints, smoothing the trajectories, and minimizing errors with demonstrations. This "gray box" method marries the optimization-based safety and interpretability with the powerful representational abilities of neural networks. We quantitatively evaluate LeTO in simulation and on the real robot. In simulation, LeTO achieves a success rate comparable to state-of-the-art imitation learning methods, but the generated trajectories are of less uncertainty, higher quality, and smoother. In real-world experiments, we deployed LeTO to handle constraints-critical tasks. The results show the effectiveness of LeTO comparing with state-of-the-art imitation learning approaches. We release our code at https://github.com/ZhengtongXu/LeTO.
Reliable offroad autonomy requires low-latency, high-accuracy state estimates of pose as well as velocity, which remain viable throughout environments with sub-optimal operating conditions for the utilized perception modalities. As state estimation remains a single point of failure system in the majority of aspiring autonomous systems, failing to address the environmental degradation the perception sensors could potentially experience given the operating conditions, can be a mission-critical shortcoming. In this work, a method for integration of radar velocity information in a LiDAR-inertial odometry solution is proposed, enabling consistent estimation performance even with degraded LiDAR-inertial odometry. The proposed method utilizes the direct velocity-measuring capabilities of an Frequency Modulated Continuous Wave (FMCW) radar sensor to enhance the LiDAR-inertial smoother solution onboard the vehicle through integration of the forward velocity measurement into the graph-based smoother. This leads to increased robustness in the overall estimation solution, even in the absence of LiDAR data. This method was validated by hardware experiments conducted onboard an all-terrain vehicle traveling at high speed, ~12 m/s, in demanding offroad environments.
This paper addresses the problem of autonomous robotic inspection in complex and unknown environments. This capability is crucial for efficient and precise inspections in various real-world scenarios, even when faced with perceptual uncertainty and lack of prior knowledge of the environment. Existing methods for real-world autonomous inspections typically rely on predefined targets and waypoints and often fail to adapt to dynamic or unknown settings. In this work, we introduce the Semantic Belief Behavior Graph (SB2G) framework as a novel approach to semantic-aware autonomous robot inspection. SB2G generates a control policy for the robot, featuring behavior nodes that encapsulate various semantic-based policies designed for inspecting different classes of objects. We design an active semantic search behavior to guide the robot in locating objects for inspection while reducing semantic information uncertainty. The edges in the SB2G encode transitions between these behaviors. We validate our approach through simulation and real-world urban inspections using a legged robotic platform. Our results show that SB2G enables a more efficient inspection policy, exhibiting performance comparable to human-operated inspections.
Data generation is a data augmentation technique for enhancing the generalization ability for skeleton-based human action recognition. Most existing data generation methods face challenges to ensure the temporal consistency of the dynamic information for action. In addition, the data generated by these methods lack diversity when only a few training samples are available. To solve those problems, We propose a novel active generative network (AGN), which can adaptively learn various action categories by motion style transfer to generate new actions when the data for a particular action is only a single sample or few samples. The AGN consists of an action generation network and an uncertainty metric network. The former, with ST-GCN as the Backbone, can implicitly learn the morphological features of the target action while preserving the category features of the source action. The latter guides generating actions. Specifically, an action recognition model generates prediction vectors for each action, which is then scored using an uncertainty metric. Finally, UMN provides the uncertainty sampling basis for the generated actions.