The Impact of Task Underspecification in Evaluating Deep Reinforcement Learning

Evaluations of Deep Reinforcement Learning (DRL) methods are an integral part of scientific progress of the field. Beyond designing DRL methods for general intelligence, designing task-specific methods is becoming increasingly prominent for real-world applications. In these settings, the standard evaluation practice involves using a few instances of Markov Decision Processes (MDPs) to represent the task. However, many tasks induce a large family of MDPs owing to variations in the underlying environment, particularly in real-world contexts. For example, in traffic signal control, variations may stem from intersection geometries and traffic flow levels. The select MDP instances may thus inadvertently cause overfitting, lacking the statistical power to draw conclusions about the method's true performance across the family. In this article, we augment DRL evaluations to consider parameterized families of MDPs. We show that in comparison to evaluating DRL methods on select MDP instances, evaluating the MDP family often yields a substantially different relative ranking of methods, casting doubt on what methods should be considered state-of-the-art. We validate this phenomenon in standard control benchmarks and the real-world application of traffic signal control. At the same time, we show that accurately evaluating on an MDP family is nontrivial. Overall, this work identifies new challenges for empirical rigor in reinforcement learning, especially as the outcomes of DRL trickle into downstream decision-making.

Automatic Identification and Classification of Share Buybacks and their Effect on Short-, Mid- and Long-Term Returns

This thesis investigates share buybacks, specifically share buyback announcements. It addresses how to recognize such announcements, the excess return of share buybacks, and the prediction of returns after a share buyback announcement. We illustrate two NLP approaches for the automated detection of share buyback announcements. Even with very small amounts of training data, we can achieve an accuracy of up to 90%. This thesis utilizes these NLP methods to generate a large dataset consisting of 57,155 share buyback announcements. By analyzing this dataset, this thesis aims to show that most companies, which have a share buyback announced are underperforming the MSCI World. A minority of companies, however, significantly outperform the MSCI World. This significant overperformance leads to a net gain when looking at the averages of all companies. If the benchmark index is adjusted for the respective size of the companies, the average overperformance disappears, and the majority underperforms even greater. However, it was found that companies that announce a share buyback with a volume of at least 1% of their market cap, deliver, on average, a significant overperformance, even when using an adjusted benchmark. It was also found that companies that announce share buybacks in times of crisis emerge better than the overall market. Additionally, the generated dataset was used to train 72 machine learning models. Through this, it was able to find many strategies that could achieve an accuracy of up to 77% and generate great excess returns. A variety of performance indicators could be improved across six different time frames and a significant overperformance was identified. This was achieved by training several models for different tasks and time frames as well as combining these different models, generating significant improvement by fusing weak learners, in order to create one strong learner.

Application of the nnU-Net for automatic segmentation of lung lesion on CT images, and implication on radiomic models

Lesion segmentation is a crucial step of the radiomic workflow. Manual segmentation requires long execution time and is prone to variability, impairing the realisation of radiomic studies and their robustness. In this study, a deep-learning automatic segmentation method was applied on computed tomography images of non-small-cell lung cancer patients. The use of manual vs automatic segmentation in the performance of survival radiomic models was assessed, as well. METHODS A total of 899 NSCLC patients were included (2 proprietary: A and B, 1 public datasets: C). Automatic segmentation of lung lesions was performed by training a previously developed architecture, the nnU-Net, including 2D, 3D and cascade approaches. The quality of automatic segmentation was evaluated with DICE coefficient, considering manual contours as reference. The impact of automatic segmentation on the performance of a radiomic model for patient survival was explored by extracting radiomic hand-crafted and deep-learning features from manual and automatic contours of dataset A, and feeding different machine learning algorithms to classify survival above/below median. Models' accuracies were assessed and compared. RESULTS The best agreement between automatic and manual contours with DICE=0.78 +(0.12) was achieved by averaging predictions from 2D and 3D models, and applying a post-processing technique to extract the maximum connected component. No statistical differences were observed in the performances of survival models when using manual or automatic contours, hand-crafted, or deep features. The best classifier showed an accuracy between 0.65 and 0.78. CONCLUSION The promising role of nnU-Net for automatic segmentation of lung lesions was confirmed, dramatically reducing the time-consuming physicians' workload without impairing the accuracy of survival predictive models based on radiomics.

TEFL: Turbo Explainable Federated Learning for 6G Trustworthy Zero-Touch Network Slicing

Sixth-generation (6G) networks anticipate intelligently supporting a massive number of coexisting and heterogeneous slices associated with various vertical use cases. Such a context urges the adoption of artificial intelligence (AI)-driven zero-touch management and orchestration (MANO) of the end-to-end (E2E) slices under stringent service level agreements (SLAs). Specifically, the trustworthiness of the AI black-boxes in real deployment can be achieved by explainable AI (XAI) tools to build transparency between the interacting actors in the slicing ecosystem, such as tenants, infrastructure providers and operators. Inspired by the turbo principle, this paper presents a novel iterative explainable federated learning (FL) approach where a constrained resource allocation model and an \emph{explainer} exchange -- in a closed loop (CL) fashion -- soft attributions of the features as well as inference predictions to achieve a transparent and SLA-aware zero-touch service management (ZSM) of 6G network slices at RAN-Edge setup under non-independent identically distributed (non-IID) datasets. In particular, we quantitatively validate the faithfulness of the explanations via the so-called attribution-based \emph{confidence metric} that is included as a constraint in the run-time FL optimization task. In this respect, Integrated-Gradient (IG) as well as Input $\times$ Gradient and SHAP are used to generate the attributions for the turbo explainable FL (TEFL), wherefore simulation results under different methods confirm its superiority over an unconstrained Integrated-Gradient \emph{post-hoc} FL baseline.

Generalizing in the Real World with Representation Learning

Machine learning (ML) formalizes the problem of getting computers to learn from experience as optimization of performance according to some metric(s) on a set of data examples. This is in contrast to requiring behaviour specified in advance (e.g. by hard-coded rules). Formalization of this problem has enabled great progress in many applications with large real-world impact, including translation, speech recognition, self-driving cars, and drug discovery. But practical instantiations of this formalism make many assumptions - for example, that data are i.i.d.: independent and identically distributed - whose soundness is seldom investigated. And in making great progress in such a short time, the field has developed many norms and ad-hoc standards, focused on a relatively small range of problem settings. As applications of ML, particularly in artificial intelligence (AI) systems, become more pervasive in the real world, we need to critically examine these assumptions, norms, and problem settings, as well as the methods that have become de-facto standards. There is much we still do not understand about how and why deep networks trained with stochastic gradient descent are able to generalize as well as they do, why they fail when they do, and how they will perform on out-of-distribution data. In this thesis I cover some of my work towards better understanding deep net generalization, identify several ways assumptions and problem settings fail to generalize to the real world, and propose ways to address those failures in practice.

Hidet: Task Mapping Programming Paradigm for Deep Learning Tensor Programs

As deep learning models nowadays are widely adopted by both cloud services and edge devices, the latency of deep learning model inferences becomes crucial to provide efficient model serving. However, it is challenging to develop efficient tensor programs for deep learning operators due to the high complexity of modern accelerators (e.g., NVIDIA GPUs and Google TPUs) and the rapidly growing number of operators. Deep learning compilers, such as Apache TVM, adopt declarative scheduling primitives to lower the bar of developing tensor programs. However, we show that this approach is insufficient to cover state-of-the-art tensor program optimizations (e.g., double buffering). In this paper, we propose to embed the scheduling process into tensor programs and use dedicated mappings, called task mappings, to define the computation assignment and ordering directly in the tensor programs. This new approach greatly enriches the expressible optimizations by allowing developers to manipulate tensor programs at a much finer granularity (e.g., allowing program statement-level optimizations). We call the proposed method the task-mapping-oriented programming paradigm. With the proposed paradigm, we implement a deep learning compiler - Hidet. Extensive experiments on modern convolution and transformer models show that Hidet outperforms state-of-the-art DNN inference framework, ONNX Runtime, and compiler, TVM equipped with scheduler AutoTVM and Ansor, by up to 1.48x (1.22x on average) with enriched optimizations. It also reduces the tuning time by 20x and 11x compared with AutoTVM and Ansor, respectively.

Research of an optimization model for servicing a network of ATMs and information payment terminals

The steadily high demand for cash contributes to the expansion of the network of Bank payment terminals. To optimize the amount of cash in payment terminals, it is necessary to minimize the cost of servicing them and ensure that there are no excess funds in the network. The purpose of this work is to create a cash management system in the network of payment terminals. The article discusses the solution to the problem of determining the optimal amount of funds to be loaded into the terminals, and the effective frequency of collection, which allows to get additional income by investing the released funds. The paper presents the results of predicting daily cash withdrawals at ATMs using a triple exponential smoothing model, a recurrent neural network with long short-term memory, and a model of singular spectrum analysis. These forecasting models allowed us to obtain a sufficient level of correct forecasts with good accuracy and completeness. The results of forecasting cash withdrawals were used to build a discrete optimal control model, which was used to develop an optimal schedule for adding funds to the payment terminal. It is proved that the efficiency and reliability of the proposed model is higher than that of the classical Baumol-Tobin inventory management model: when tested on the time series of three ATMs, the discrete optimal control model did not allow exhaustion of funds and allowed to earn on average 30% more than the classical model.

Virtual Reality via Object Poses and Active Learning: Realizing Telepresence Robots with Aerial Manipulation Capabilities

This article presents a novel telepresence system for advancing aerial manipulation in dynamic and unstructured environments. The proposed system not only features a haptic device, but also a virtual reality (VR) interface that provides real-time 3D displays of the robot's workspace as well as a haptic guidance to its remotely located operator. To realize this, multiple sensors namely a LiDAR, cameras and IMUs are utilized. For processing of the acquired sensory data, pose estimation pipelines are devised for industrial objects of both known and unknown geometries. We further propose an active learning pipeline in order to increase the sample efficiency of a pipeline component that relies on Deep Neural Networks (DNNs) based object detection. All these algorithms jointly address various challenges encountered during the execution of perception tasks in industrial scenarios. In the experiments, exhaustive ablation studies are provided to validate the proposed pipelines. Methodologically, these results commonly suggest how an awareness of the algorithms' own failures and uncertainty ("introspection") can be used tackle the encountered problems. Moreover, outdoor experiments are conducted to evaluate the effectiveness of the overall system in enhancing aerial manipulation capabilities. In particular, with flight campaigns over days and nights, from spring to winter, and with different users and locations, we demonstrate over 70 robust executions of pick-and-place, force application and peg-in-hole tasks with the DLR cable-Suspended Aerial Manipulator (SAM). As a result, we show the viability of the proposed system in future industrial applications.

Robust Time Series Denoising with Learnable Wavelet Packet Transform

In many applications, signal denoising is often the first pre-processing step before any subsequent analysis or learning task. In this paper, we propose to apply a deep learning denoising model inspired by a signal processing, a learnable version of wavelet packet transform. The proposed algorithm has signficant learning capabilities with few interpretable parameters and has an intuitive initialisation. We propose a post-learning modification of the parameters to adapt the denoising to different noise levels. We evaluate the performance of the proposed methodology on two case studies and compare it to other state of the art approaches, including wavelet schrinkage denoising, convolutional neural network, autoencoder and U-net deep models. The first case study is based on designed functions that have typically been used to study denoising properties of the algorithms. The second case study is an audio background removal task. We demonstrate how the proposed algorithm relates to the universality of signal processing methods and the learning capabilities of deep learning approaches. In particular, we evaluate the obtained denoising performances on structured noisy signals inside and outside the classes used for training. In addition to having good performance in denoising signals inside and outside to the training class, our method shows to be particularly robust when different noise levels, noise types and artifacts are added.

Prototypical VoteNet for Few-Shot 3D Point Cloud Object Detection

Most existing 3D point cloud object detection approaches heavily rely on large amounts of labeled training data. However, the labeling process is costly and time-consuming. This paper considers few-shot 3D point cloud object detection, where only a few annotated samples of novel classes are needed with abundant samples of base classes. To this end, we propose Prototypical VoteNet to recognize and localize novel instances, which incorporates two new modules: Prototypical Vote Module (PVM) and Prototypical Head Module (PHM). Specifically, as the 3D basic geometric structures can be shared among categories, PVM is designed to leverage class-agnostic geometric prototypes, which are learned from base classes, to refine local features of novel categories.Then PHM is proposed to utilize class prototypes to enhance the global feature of each object, facilitating subsequent object localization and classification, which is trained by the episodic training strategy. To evaluate the model in this new setting, we contribute two new benchmark datasets, FS-ScanNet and FS-SUNRGBD. We conduct extensive experiments to demonstrate the effectiveness of Prototypical VoteNet, and our proposed method shows significant and consistent improvements compared to baselines on two benchmark datasets.