Computing Research Repository

Jérémie, Jean-Nicolas;Perrinet, Laurent U

doi: 10.3390/vision7020029pmid: 37092462

Abstract: Humans are able to categorize images very efficiently, in particular to detect the presence of an animal very quickly. Recently, deep learning algorithms based on convolutional neural networks (CNNs) have achieved higher than human accuracy for a wide range of visual categorization tasks. However, the tasks on which these artificial networks are typically trained and evaluated tend to be highly specialized and do not generalize well, e.g., accuracy drops after image rotation. In this respect, biological visual systems are more flexible and efficient than artificial systems for more general tasks, such as recognizing an animal. To further the comparison between biological and artificial neural networks, we re-trained the standard VGG 16 CNN on two independent tasks that are ecologically relevant to humans: detecting the presence of an animal or an artifact. We show that re-training the network achieves a human-like level of performance, comparable to that reported in psychophysical tasks. In addition, we show that the categorization is better when the outputs of the models are combined. Indeed, animals (e.g., lions) tend to be less present in photographs that contain artifacts (e.g., buildings). Furthermore, these re-trained models were able to reproduce some unexpected behavioral observations from human psychophysics, such as robustness to rotation (e.g., an upside-down or tilted image) or to a grayscale transformation. Finally, we quantified the number of CNN layers required to achieve such performance and showed that good accuracy for ultrafast image categorization can be achieved with only a few layers, challenging the belief that image recognition requires deep sequential analysis of visual objects.

Kolonin, Anton;Ramesh, Vignav

doi: 10.48550/arxiv.2205.11443pmid: N/A

Abstract: In the presented study, we discover that the so-called "transition freedom" metric appears superior for unsupervised tokenization purposes in comparison to statistical metrics such as mutual information and conditional probability, providing F-measure scores in range from 0.71 to 1.0 across explored multilingual corpora. We find that different languages require different offshoots of that metric (such as derivative, variance, and "peak values") for successful tokenization. Larger training corpora do not necessarily result in better tokenization quality, while compressing the models by eliminating statistically weak evidence tends to improve performance. The proposed unsupervised tokenization technique provides quality better than or comparable to lexicon-based ones, depending on the language.

Cagiltay, Bengisu;White, Nathan;Ibtasar, Rabia;Mutlu, Bilge;Michaelis, Joseph

doi: 10.1145/3501712.3529747pmid: N/A

Abstract: Social robots are emerging as learning companions for children, and research shows that they facilitate the development of interest and learning even through brief interactions. However, little is known about how such technologies might support these goals in authentic environments over long-term periods of use and interaction. We designed a learning companion robot capable of supporting children reading popular-science books by expressing social and informational commentaries. We deployed the robot in homes of 14 families with children aged 10-12 for four weeks during the summer. Our analysis revealed critical factors that affected children's long-term engagement and adoption of the robot, including external factors such as vacations, family visits, and extracurricular activities; family/parental involvement; and children's individual interests. We present four in-depth cases that illustrate these factors and demonstrate their impact on children's reading experiences and discuss the implications of our findings for robot design.

Kim, Donghyeon;Kim, Gwantae;Lee, Bokyeung;Kwak, Jeong-gi;Han, David K.;Ko, Hanseok

doi: 10.48550/arxiv.2205.01304pmid: N/A

Abstract: Unseen noise signal which is not considered in a model training process is difficult to anticipate and would lead to performance degradation. Various methods have been investigated to mitigate unseen noise. In our previous work, an Instance-level Dynamic Filter (IDF) and a Pixel Dynamic Filter (PDF) were proposed to extract noise-robust features. However, the performance of the dynamic filter might be degraded since simple feature pooling is used to reduce the computational resource in the IDF part. In this paper, we propose an efficient dynamic filter to enhance the performance of the dynamic filter. Instead of utilizing the simple feature mean, we separate Time-Frequency (T-F) features as non-overlapping chunks, and separable convolutions are carried out for each feature direction (inter chunks and intra chunks). Additionally, we propose Dynamic Attention Pooling that maps high dimensional features as low dimensional feature embeddings. These methods are applied to the IDF for keyword spotting and speaker verification tasks. We confirm that our proposed method performs better in unseen environments (unseen noise and unseen speakers) than state-of-the-art models.

Papachristou, Marios;Banerjee, Siddhartha;Kleinberg, Jon

doi: 10.1145/3543507.3583470pmid: N/A

Abstract: We study the problem of designing dynamic intervention policies for minimizing networked defaults in financial networks. Formally, we consider a dynamic version of the celebrated Eisenberg-Noe model of financial network liabilities and use this to study the design of external intervention policies. Our controller has a fixed resource budget in each round and can use this to minimize the effect of demand/supply shocks in the network. We formulate the optimal intervention problem as a Markov Decision Process and show how we can leverage the problem structure to efficiently compute optimal intervention policies with continuous interventions and provide approximation algorithms for discrete interventions. Going beyond financial networks, we argue that our model captures dynamic network intervention in a much broader class of dynamic demand/supply settings with networked inter-dependencies. To demonstrate this, we apply our intervention algorithms to various application domains, including ridesharing, online transaction platforms, and financial networks with agent mobility. In each case, we study the relationship between node centrality and intervention strength, as well as the fairness properties of the optimal interventions.

Vacková, Jana;Bukáček, Marek

doi: 10.1080/23249935.2023.2236236pmid: N/A

Abstract: The standard definition of pedestrian density produces scattered values, hence, many approaches have been developed to improve the features of the estimated density. This paper provides a review of generally applied methods and presents a general framework based on various kernels that bring desired properties of density estimates (e.g., continuity) and incorporate ordinarily used methods. The developed kernel concept considers each pedestrian as a source of density distribution, parametrized by the kernel type (e.g., Gauss, cone) and kernel size. The quantitative parametric study performed on experimental data illustrates that parametrization brings desired features, for instance, a conic kernel with a base radius in (0.7, 1.2) m produces smooth values that retain trend features. The correspondence between kernel and non-kernel methods (namely Voronoi diagram and customized inverse distance to the nearest pedestrian) is achievable for a wide range of kernel parameter. Thereby the generality of the concept is supported.

Furui, Kairi;Ohue, Masahito

doi: 10.1109/cibcb55180.2022.9863032pmid: N/A

Abstract: Learning-to-rank, a machine learning technique widely used in information retrieval, has recently been applied to the problem of ligand-based virtual screening, to accelerate the early stages of new drug development. Ranking prediction models learn based on ordinal relationships, making them suitable for integrating assay data from various environments. Existing studies of rank prediction in compound screening have generally used a learning-to-rank method called RankSVM. However, they have not been compared with or validated against the gradient boosting decision tree (GBDT)-based learning-to-rank methods that have gained popularity recently. Furthermore, although the ranking metric called Normalized Discounted Cumulative Gain (NDCG) is widely used in information retrieval, it only determines whether the predictions are better than those of other models. In other words, NDCG is incapable of recognizing when a prediction model produces worse than random results. Nevertheless, NDCG is still used in the performance evaluation of compound screening using learning-to-rank. This study used the GBDT model with ranking loss functions, called lambdarank and lambdaloss, for ligand-based virtual screening; results were compared with existing RankSVM methods and GBDT models using regression. We also proposed a new ranking metric, Normalized Enrichment Discounted Cumulative Gain (NEDCG), which aims to properly evaluate the goodness of ranking predictions. Results showed that the GBDT model with learning-to-rank outperformed existing regression methods using GBDT and RankSVM on diverse datasets. Moreover, NEDCG showed that predictions by regression were comparable to random predictions in multi-assay, multi-family datasets, demonstrating its usefulness for a more direct assessment of compound screening performance.

Arribas, Edgar;Cholvi, Vicent;Mancuso, Vincenzo

doi: 10.1109/tro.2023.3263077pmid: N/A

Abstract:The adoption of UAVs in communication networks is becoming reality thanks to the deployment of advanced solutions for connecting UAVs and using them as communication relays. However, the use of UAVs introduces novel energy constraints and scheduling challenges in the dynamic management of network devices, due to the need to call back and recharge, or substitute, UAVs that run out of energy. In this paper, we design UAV recharging schemes under realistic assumptions on limited flight times and time consuming charging operations. Such schemes are designed to minimize the size of the fleet to be devoted to a persistent service of a set of aerial locations, hence its cost. We consider a fleet of homogeneous UAVs both under homogeneous and heterogeneous service topologies. For UAVs serving aerial locations with homogeneous distances to a recharge station, we design a simple scheduling, that we name HORR, which we prove to be feasible and optimal, in the sense that it uses the minimum possible number of UAVs to guarantee the coverage of the aerial service locations. For the case of non-evenly distributed aerial locations, we demonstrate that the problem becomes NP-hard, and design a lightweight recharging scheduling scheme, PHERR, that extends the operation of HORR to the heterogeneous case, leveraging the partitioning of the set of service locations. We show that PHERR is near-optimal because it approaches the performance limits identified through a lower bound that we formulate on the total fleet size.

Putra, Rachmad Vidya Wicaksana;Shafique, Muhammad

doi: 10.1109/ijcnn55064.2022.9892948pmid: N/A

Abstract: Recent advances have shown that SNN-based systems can efficiently perform unsupervised continual learning due to their bio-plausible learning rule, e.g., Spike-Timing-Dependent Plasticity (STDP). Such learning capabilities are especially beneficial for use cases like autonomous agents (e.g., robots and UAVs) that need to continuously adapt to dynamically changing scenarios/environments, where new data gathered directly from the environment may have novel features that should be learned online. Current state-of-the-art works employ high-precision weights (i.e., 32 bit) for both training and inference phases, which pose high memory and energy costs thereby hindering efficient embedded implementations of such systems for battery-driven mobile autonomous systems. On the other hand, precision reduction may jeopardize the quality of unsupervised continual learning due to information loss. Towards this, we propose lpSpikeCon, a novel methodology to enable low-precision SNN processing for efficient unsupervised continual learning on resource-constrained autonomous agents/systems. Our lpSpikeCon methodology employs the following key steps: (1) analyzing the impacts of training the SNN model under unsupervised continual learning settings with reduced weight precision on the inference accuracy; (2) leveraging this study to identify SNN parameters that have a significant impact on the inference accuracy; and (3) developing an algorithm for searching the respective SNN parameter values that improve the quality of unsupervised continual learning. The experimental results show that our lpSpikeCon can reduce weight memory of the SNN model by 8x (i.e., by judiciously employing 4-bit weights) for performing online training with unsupervised continual learning and achieve no accuracy loss in the inference phase, as compared to the baseline model with 32-bit weights across different network sizes.

Rosenfelder, Mario;Ebel, Henrik;Krauspenhaar, Jasmin;Eberhard, Peter

doi: 10.1016/j.automatica.2023.110972pmid: N/A

Abstract: Non-holonomic vehicles are of immense practical value and increasingly subject to automation. However, controlling them accurately, e.g., when parking, is known to be challenging for automatic control methods, including model predictive control (MPC). Combining results from MPC theory and sub-Riemannian geometry in the form of homogeneous nilpotent system approximations, this paper proposes a comprehensive, ready-to-apply design procedure for MPC controllers to steer controllable, driftless non-holonomic vehicles into given setpoints. It can be ascertained that the resulting controllers nominally asymptotically stabilize the setpoint for a large-enough prediction horizon. The design procedure is exemplarily applied to four vehicles, including the kinematic car and a differentially driven mobile robot with up to two trailers. The controllers use a non-quadratic cost function tailored to the non-holonomic kinematics. Novelly, for the considered example vehicles, it is proven that a quadratic cost employed in an otherwise similar controller is insufficient to reliably asymptotically stabilize the closed loop. Since quadratic costs are the conventional choice in control, this highlights the relevance of the findings. To the knowledge of the authors, it is the first time that MPC controllers of the proposed structure are applied to non-holonomic vehicles beyond very simple ones, in particular (partly) on hardware.