A key challenge in building theoretical foundations for deep learning is the complex optimization dynamics of neural networks, resulting from the high-dimensional interactions between the large number of network parameters. Such non-trivial dynamics lead to intriguing behaviors such as the phenomenon of "double descent" of the generalization error. The more commonly studied aspect of this phenomenon corresponds to model-wise double descent where the test error exhibits a second descent with increasing model complexity, beyond the classical U-shaped error curve. In this work, we investigate the origins of the less studied epoch-wise double descent in which the test error undergoes two non-monotonous transitions, or descents as the training time increases. By leveraging tools from statistical physics, we study a linear teacher-student setup exhibiting epoch-wise double descent similar to that in deep neural networks. In this setting, we derive closed-form analytical expressions for the evolution of generalization error over training. We find that double descent can be attributed to distinct features being learned at different scales: as fast-learning features overfit, slower-learning features start to fit, resulting in a second descent in test error. We validate our findings through numerical experiments where our theory accurately predicts empirical findings and remains consistent with observations in deep neural networks.
3D visual tracking is significant to deep space exploration programs, which can guarantee spacecraft to flexibly approach the target. In this paper, we focus on the studied accurate and real-time method for 3D tracking. Considering the fact that there are almost no public dataset for this topic, A new large-scale 3D asteroid tracking dataset is presented, including binocular video sequences, depth maps, and point clouds of diverse asteroids with various shapes and textures. Benefitting from the power and convenience of simulation platform, all the 2D and 3D annotations are automatically generated. Meanwhile, we propose a deep-learning based 3D tracking framework, named as Track3D, which involves 2D monocular tracker and a novel light-weight amodal axis-aligned bounding-box network, A3BoxNet. The evaluation results demonstrate that Track3D achieves state-of-the-art 3D tracking performance in both accuracy and precision, comparing to a baseline algorithm. Moreover, our framework has great generalization ability to 2D monocular tracking performance.
Tree skeleton plays an important role in tree structure analysis, forest inventory and ecosystem monitoring. However, it is a challenge to extract a skeleton from a tree point cloud with complex branches. In this paper, an automatic and fast tree skeleton extraction method (FTSEM) based on voxel thinning is proposed. In this method, a wood-leaf classification algorithm was introduced to filter leaf points for the reduction of the leaf interference on tree skeleton generation, tree voxel thinning was adopted to extract raw tree skeleton quickly, and a breakpoint connection algorithm was used to improve the skeleton connectivity and completeness. Experiments were carried out in Haidian Park, Beijing, in which 24 trees were scanned and processed to obtain tree skeletons. The graph search algorithm (GSA) is used to extract tree skeletons based on the same datasets. Compared with GSA method, the FTSEM method obtained more complete tree skeletons. And the time cost of the FTSEM method is evaluated using the runtime and time per million points (TPMP). The runtime of FTSEM is from 1.0 s to 13.0 s, and the runtime of GSA is from 6.4 s to 309.3 s. The average value of TPMP is 1.8 s for FTSEM, and 22.3 s for GSA respectively. The experimental results demonstrate that the proposed method is feasible, robust, and fast with a good potential on tree skeleton extraction.
It has been shown that the apparent advantage of some quantum machine learning algorithms may be efficiently replicated using classical algorithms with suitable data access -- a process known as dequantization. Existing works on dequantization compare quantum algorithms which take copies of an n-qubit quantum state $|x\rangle = \sum_{i} x_i |i\rangle$ as input to classical algorithms which have sample and query (SQ) access to the vector $x$. In this note, we prove that classical algorithms with SQ access can accomplish some learning tasks exponentially faster than quantum algorithms with quantum state inputs. Because classical algorithms are a subset of quantum algorithms, this demonstrates that SQ access can sometimes be significantly more powerful than quantum state inputs. Our findings suggest that the absence of exponential quantum advantage in some learning tasks may be due to SQ access being too powerful relative to quantum state inputs. If we compare quantum algorithms with quantum state inputs to classical algorithms with access to measurement data on quantum states, the landscape of quantum advantage can be dramatically different. We remark that when the quantum states are constructed from exponential-size classical data, comparing SQ access and quantum state inputs is appropriate since both require exponential time to prepare.
Multi-task learning is a powerful method for solving several tasks jointly by learning robust representation. Optimization of the multi-task learning model is a more complex task than a single-task due to task conflict. Based on theoretical results, convergence to the optimal point is guaranteed when step size is chosen through line search. But, usually, line search for the step size is not the best choice due to the large computational time overhead. We propose a novel idea for line search algorithms in multi-task learning. The idea is to use latent representation space instead of parameter space for finding step size. We examined this idea with backtracking line search. We compare this fast backtracking algorithm with classical backtracking and gradient methods with a constant learning rate on MNIST, CIFAR-10, Cityscapes tasks. The systematic empirical study showed that the proposed method leads to more accurate and fast solution, than the traditional backtracking approach and keep competitive computational time and performance compared to the constant learning rate method.
A classification technique incorporating a novel feature derivation method is proposed for predicting failure of a system or device with multivariate time series sensor data. We treat the multivariate time series sensor data as images for both visualization and computation. Failure follows various patterns which are closely related to the root causes. Different predefined transformations are applied on the original sensors data to better characterize the failure patterns. In addition to feature derivation, ensemble method is used to further improve the performance. In addition, a general algorithm architecture of deep neural network is proposed to handle multiple types of data with less manual feature engineering. We apply the proposed method on the early predict failure of computer disk drive in order to improve storage systems availability and avoid data loss. The classification accuracy is largely improved with the enriched features, named smart features.
Cell-free massive multiple-input multiple-output (MIMO) employs a large number of distributed access points (APs) to serve a small number of user equipments (UEs) via the same time/frequency resource. Due to the strong macro diversity gain, cell-free massive MIMO can considerably improve the achievable sum-rate compared to conventional cellular massive MIMO. However, the performance of cell-free massive MIMO is upper limited by inter-user interference (IUI) when employing simple maximum ratio combining (MRC) at receivers. To harness IUI, the expanded compute-and-forward (ECF) framework is adopted. In particular, we propose power control algorithms for the parallel computation and successive computation in the ECF framework, respectively, to exploit the performance gain and then improve the system performance. Furthermore, we propose an AP selection scheme and the application of different decoding orders for the successive computation. Finally, numerical results demonstrate that ECF frameworks outperform the conventional CF and MRC frameworks in terms of achievable sum-rate.
Indoor positioning is one of the core technologies of Internet of Things (IoT) and artificial intelligence (AI), and is expected to play a significant role in the upcoming era of AI. However, affected by the complexity of indoor environments, it is still highly challenging to achieve continuous and reliable indoor positioning. Currently, 5G cellular networks are being deployed worldwide, the new technologies of which have brought the approaches for improving the performance of wireless indoor positioning. In this paper, we investigate the indoor positioning under the 5G new radio (NR), which has been standardized and being commercially operated in massive markets. Specifically, a solution is proposed and a software defined receiver (SDR) is developed for indoor positioning. With our SDR indoor positioning system, the 5G NR signals are firstly sampled by universal software radio peripheral (USRP), and then, coarse synchronization is achieved via detecting the start of the synchronization signal block (SSB). Then, with the assistance of the pilots transmitted on the physical broadcasting channel (PBCH), multipath acquisition and delay tracking are sequentially carried out to estimate the time of arrival (ToA) of received signals. Furthermore, to improve the ToA ranging accuracy, the carrier phase of the first arrived path is estimated. Finally, to quantify the accuracy of our ToA estimation method, indoor field tests are carried out in an office environment, where a 5G NR base station (known as gNB) is installed for commercial use. Our test results show that, in the static test scenarios, the ToA accuracy measured by the 1-{\sigma} error interval is about 0.5 m, while in the pedestrian mobile environment, the probability of range accuracy within 0.8 m is 95%.
Many real-world vehicle routing problems involve rich sets of constraints with respect to the capacities of the vehicles, time windows for customers etc. While in recent years first machine learning models have been developed to solve basic vehicle routing problems faster than optimization heuristics, complex constraints rarely are taken into consideration. Due to their general procedure to construct solutions sequentially route by route, these methods generalize unfavorably to such problems. In this paper, we develop a policy model that is able to start and extend multiple routes concurrently by using attention on the joint action space of several tours. In that way the model is able to select routes and customers and thus learns to make difficult trade-offs between routes. In comprehensive experiments on three variants of the vehicle routing problem with time windows we show that our model called JAMPR works well for different problem sizes and outperforms the existing state-of-the-art constructive model. For two of the three variants it also creates significantly better solutions than a comparable meta-heuristic solver.
Tracking multiple time-varying states based on heterogeneous observations is a key problem in many applications. Here, we develop a statistical model and algorithm for tracking an unknown number of targets based on the probabilistic fusion of observations from two classes of data sources. The first class, referred to as target-independent perception systems (TIPSs), consists of sensors that periodically produce noisy measurements of targets without requiring target cooperation. The second class, referred to as target-dependent reporting systems (TDRSs), relies on cooperative targets that report noisy measurements of their state and their identity. We present a joint TIPS-TDRS observation model that accounts for observation-origin uncertainty, missed detections, false alarms, and asynchronicity. We then establish a factor graph that represents this observation model along with a state evolution model including target identities. Finally, by executing the sum-product algorithm on that factor graph, we obtain a scalable multitarget tracking algorithm with inherent TIPS-TDRS fusion. The performance of the proposed algorithm is evaluated using simulated data as well as real data from a maritime surveillance experiment.