Large language models (LLMs) have developed impressive performance and strong explainability across various reasoning scenarios, marking a significant stride towards mimicking human-like intelligence. Despite this, when tasked with several simple questions supported by a generic fact, LLMs often struggle to abstract and apply the generic fact to provide consistent and precise answers, revealing a deficiency in abstract reasoning abilities. This has sparked a vigorous debate about whether LLMs are genuinely reasoning or merely memorizing. In light of this, we design a preliminary study to quantify and delve into the abstract reasoning abilities of existing LLMs. Our findings reveal a substantial discrepancy between their general reasoning and abstract reasoning performances. To relieve this problem, we tailor an abstract reasoning dataset (AbsR) together with a meaningful learning paradigm to teach LLMs how to leverage generic facts for reasoning purposes. The results show that our approach not only boosts the general reasoning performance of LLMs but also makes considerable strides towards their capacity for abstract reasoning, moving beyond simple memorization or imitation to a more nuanced understanding and application of generic facts.

The problem of monotone submodular maximization has been studied extensively due to its wide range of applications. However, there are cases where one can only access the objective function in a distorted or noisy form because of the uncertain nature or the errors involved in the evaluation. This paper considers the problem of constrained monotone submodular maximization with noisy oracles introduced by [11].

High-quality public datasets significantly prompt the prosperity of deep neural networks (DNNs).

Bayesian optimization is an effective technique for black-box optimization, but its applicability is typically limited to low-dimensional and small-budget problems due to the cubic complexity of computing the Gaussian process (GP) surrogate. While various approximate GP models have been employed to scale Bayesian optimization to larger sample sizes, most suffer from overlysmooth estimation and focus primarily on problems that allow for large online samples. In this work, we argue that Bayesian optimization algorithms with sparse GPs can more efficiently allocate their representational power to relevant regions of the search space. To achieve this, we propose focalized GP, which leverages a novel variational loss function to achieve stronger local prediction, as well as FocalBO, which hierarchically optimizes the focalized GP acquisition function over progressively smaller search spaces. Experimental results demonstrate that FocalBO can efficiently leverage large amounts of offline and online data to achieve state-of-the-art performance on robot morphology design and to control a 585-dimensional musculoskeletal system.

Volume rendering in neural radiance fields is inherently time-consuming due to the large number of MLP calls on the points sampled per ray. Previous works would address this issue by introducing new neural networks or data structures. In this work, we propose GL-NeRF, a new perspective of computing volume rendering with the Gauss-Laguerre quadrature. GL-NeRF significantly reduces the number of MLP calls needed for volume rendering, introducing no additional data structures or neural networks. The simple formulation makes adopting GL-NeRF in any NeRF model possible. In the paper, we first justify the use of the Gauss-Laguerre quadrature and then demonstrate this plug-and-play attribute by implementing it in two different NeRF models. We show that with a minimal drop in performance, GL-NeRF can significantly reduce the number of MLP calls, showing the potential to speed up any NeRF model.

Both the dataset and the source code are released under two licenses: (1) Creative Commons CC-BY-NC 4.0 license and (2) a custom license. The users/data requestors must agree to both licenses, and it is to be noted that if there is any conflict between any term(s) between two licenses, the custom license shall take priority over the Creative Commons CC-BY-NC 4.0 license. Each session per participant in the conventional lab setting consists of four recordings, each lasting approximately four minutes. In the first two recordings, the participants wore the DAVIS346 camera whereas in the last two recordings, they wore the Pupil-Core eye tracker. The randomized movement pattern of the visual stimulus, i.e., the white circle was identical across the cross-device recording pair (for both the DAVIS346 and Pupil-Core device) but varied between the two recordings corresponding to the same wearable device. In the ambient luminance-changing settings, each session per participant (seated in an office environment similar to conventional lab settings) consists of four recordings. For the first two recordings, the participant wears the DAVIS346 sensor under two lighting conditions: Constant Lighting Condition: Near-eye Lux value maintained at 65 Lux throughout the experiment. Variable Lighting Condition: Near-eye Lux value alternates between 65 Lux and 8 Lux every 1-minute span. For the last two recordings, the participant wears the Pupil-Core eye tracker under the two lighting conditions mentioned above. The participant mobility settings mirror the ambient luminance-changing settings for data recording, however with two mobility conditions (with constant default lighting condition of near-eye 65 lux): Stationary Condition: Sitting in an office environment Mobile Condition: Moving freely while carrying a 14-inch laptop that displays the visual stimuli.

In Algorithm 1 we present the memorizing and recalling procedures of the single-layer tPC. In Algorithm 2 we present the memorizing and recalling procedures of the 2-layer tPC. It is worth noting that, although in both algorithms we used iterative inference (line 14-16 in Algorithm 1 and line 17-19 in Algorithm 2), these inferential dynamics can be replaced by forward passes in simulation. However, obtaining the retrievals via iterative methods allows us to implement the computations in the plausible neural circuits in Figure 1 whereas forward passes cannot. Code will be available upon acceptance.