Fair Benchmarking Research Articles

The Effect of Strengthening Responsible Character Education on Student Learning Achievement Through Civics Learning at SMP Negeri 8 Satap Tondano

The problem in this research is how to maintain the mandatory school personality which influences student learning achievement at SMP Negeri 8 Satap Tondano. This researcher hopes to see whether there is an influence of the required character strengthening efforts on student learning achievement. The methodology used in this test is a preliminary strategy with a Fair Benchmark Gathering Plan pretest and posttest. The population in this survey were all students of SMP Negeri 8 Satap Tondano and the investigation test was 20 students (10 students in class VIII and 10 students in class IX) who were randomly selected from the investigation class and the control class. Judging from the results of the assessment using the t-test at the real level α = 0.05, t_count = 9.952 and t_table = 1.734, because t_count > t_table then H_(o) is rejected and H_1 is recognized. Thus, it is hoped that efforts to strengthen the character of commitment can influence student learning achievement. The scientists proposed that this exploration could serve as a reference in exploring ways to further develop careful individual preparation to help further promote student achievement learning.

Revisiting Realistic Test-Time Training: Sequential Inference and Adaptation by Anchored Clustering Regularized Self-Training.

Deploying models on target domain data subject to distribution shift requires adaptation. Test-time training (TTT) emerges as a solution to this adaptation under a realistic scenario where access to full source domain data is not available, and instant inference on the target domain is required. Despite many efforts into TTT, there is a confusion over the experimental settings, thus leading to unfair comparisons. In this work, we first revisit TTT assumptions and categorize TTT protocols by two key factors, i.e., whether testing data is sequentially streamed and whether source model is allowed to be trained with modified loss function. Among the multiple protocols, we adopt a realistic sequential test-time training (sTTT) protocol, under which we develop a test-time anchored clustering (TTAC) approach to enable stronger test-time feature learning. TTAC discovers clusters in both source and target domains and matches the target clusters to the source ones to improve adaptation. When source domain information is strictly absent (i.e., source-free) we further develop an efficient method to infer source domain distributions for anchored clustering. Finally, self-training (ST) has demonstrated great success in learning from unlabeled data and we empirically figure out that applying ST alone to TTT is prone to confirmation bias. Therefore, a more effective TTT approach is introduced by regularizing self-training with anchored clustering, and the improved model is referred to as TTAC++. We demonstrate that, under all TTT protocols, TTAC++ consistently outperforms the state-of-the-art methods on five TTT datasets, including corrupted target domain, selected hard samples, synthetic-to-real adaptation and adversarially attacked target domain. We hope this work will provide a fair benchmarking of TTT methods, and future research should be compared within respective protocols.

Identification of challenges to Industry 5.0 adoption in Indian manufacturing firms: an emerging economy perspective

ABSTRACT This study aims to identify a comprehensive list of challenges and empirically evaluate their impact on Industry 5.0 adoption, particularly in developing countries in Indian manufacturing firms. Challenges were classified into four significant categories: human, environmental, legal and ethical, and financial through an extensive literature review conducted and experts’ inputs. Multi-criteria decision-making (MCDM) is a method for analyzing complex issues by assessing multiple alternatives based on specific criteria. The analytic hierarchy Process (AHP) aids in selecting suitable options, while the analytical network processes (ANP) technique determines relevant criteria. A sensitivity analysis was carried out to validate the research findings. The most preeminent challenges were identified as human critical thinking, human-machine teams, incorporation of ethics, investment in Industry 5.0 technologies, human well-being perspective, over-dependency on technology, and personalization and hyper-customization 5.0, respectively, in the adoption of Industry 5.0. The comprehensive challenges proposed in the study will assist decision-makers, practitioners, and managers in determining fair benchmarking strategies to precipitate Industry 5.0. This research helps to gauge manufacturing firms’ readiness index to adopt Industry 5.0, provides a distinct perspective, and foregrounds the significance of research to assist the industry in its long-term to anticipate Production 6.0.

Equal tax and equal compensation: A fair and efficient way to save climate

We show that “equal tax and equal compensation” (T&C) is fair as justified by the two fairness principles. It differs from any Pigouvian tax with fixed lump-sum payments and can motivate every country to maximize world welfare. It benefits countries with current per capita emissions lower than the world average and would benefit every country when compared with a fair benchmark where emissions are duly penalized and compensated. Subsidizing emission reduction by poll tax is Pareto efficient and Pareto improving over status quo, but unfair. An imperfect T&C with a sub-optimal tax or pyramid taxes can still benefit the world.

TFB: Towards Comprehensive and Fair Benchmarking of Time Series Forecasting Methods

Time series are generated in diverse domains such as economic, traffic, health, and energy, where forecasting of future values has numerous important applications. Not surprisingly, many forecasting methods are being proposed. To ensure progress, it is essential to be able to study and compare such methods empirically in a comprehensive and reliable manner. To achieve this, we propose TFB, an automated benchmark for Time Series Forecasting (TSF) methods. TFB advances the state-of-the-art by addressing shortcomings related to datasets, comparison methods, and evaluation pipelines: 1) insufficient coverage of data domains, 2) stereotype bias against traditional methods, and 3) inconsistent and inflexible pipelines. To achieve better domain coverage, we include datasets from 10 different domains : traffic, electricity, energy, the environment, nature, economic, stock markets, banking, health, and the web. We also provide a time series characterization to ensure that the selected datasets are comprehensive. To remove biases against some methods, we include a diverse range of methods, including statistical learning, machine learning, and deep learning methods, and we also support a variety of evaluation strategies and metrics to ensure a more comprehensive evaluations of different methods. To support the integration of different methods into the benchmark and enable fair comparisons, TFB features a flexible and scalable pipeline that eliminates biases. Next, we employ TFB to perform a thorough evaluation of 21 Univariate Time Series Forecasting (UTSF) methods on 8,068 univariate time series and 14 Multivariate Time Series Forecasting (MTSF) methods on 25 datasets. The results offer a deeper understanding of the forecasting methods, allowing us to better select the ones that are most suitable for particular datasets and settings. Overall, TFB and this evaluation provide researchers with improved means of designing new TSF methods.

Moving Towards Industry 5.0 in the Pharmaceutical Manufacturing Sector: Challenges and Solutions for Germany

Emergence of industry 5.0 facilitates real-time synchronisation of production processes, which helps with the production of customized products. Although industry 5.0 can revolutionize the landscape of goods production, still its adoption is in the infancy stage. Hence, the purpose of this article is to examine barriers that impede the adoption of industry 5.0, and to propose solution initiatives (SIs). After an in-depth literature review and interviews with German industry experts in pharmaceutical sector, barriers and SIs are ranked using an integrated Analytical hierarchy process-elimination and choice expressing reality-decision-making trial and evaluation laboratory (AHP-ELECTRE-DEMATEL) approach. “Linking virtual reality and reality” is found to be the most critical deterrent to the adoption of industry 5.0, and falls into a causal group, which signifies their influence on other deterrents. “Measures for better connectivity with patients” is of utmost importance for German firms to ensure secure communication and safeguard patient data. The findings also highlight the problems with the adoption of high-tech innovations due to lack of standardization and fair benchmarking policies on industry 5.0. This article concludes by proposing an industry 5.0 framework. The inhibitors to adopting industry 5.0 were investigated in this article, the results of which will assist practitioners and decision makers to understand the issues with industry 5.0. The results of this article will help potential adopters of this technology to look for the SIs. This will therefore pave the path for effective adoption of industry 5.0 across the pharmaceutical manufacturing sector.

Optimal Radio Propagation Modeling and Parametric Tuning Using Optimization Algorithms

Benchmarking different optimization algorithms is tasky, particularly for network-based cellular communication systems. The design and management process of these systems involves many stochastic variables and complex design parameters that demand an unbiased estimation and analysis. Though several optimization algorithms exist for different parametric modeling and tuning, an in-depth evaluation of their functional performance has not been adequately addressed, especially for cellular communication systems. Firstly, in this paper, nine key numerical and global optimization algorithms, comprising Gauss–Newton (GN), gradient descent (GD), Genetic Algorithm (GA), Levenberg–Marguardt (LM), Quasi-Newton (QN), Trust-Region–Dog-Leg (TR), pattern search (PAS), Simulated Annealing (SA), and particle swam (PS), have been benchmarked against measured data. The experimental data were taken from different radio signal propagation terrains around four eNodeB cells. In order to assist the radio frequency (RF) engineer in selecting the most suitable optimization method for the parametric model tuning, three-fold benchmarking criteria comprising the Accuracy Profile Benchmark (APB), Function Evaluation Benchmark (FEB), and Execution Speed Benchmark (ESB) were employed. The APB and FEB were quantitatively compared against the measured data for fair benchmarking. By leveraging the APB performance criteria, the QN achieved the best results with the preferred values of 98.34, 97.31, 97.44, and 96.65% in locations 1–4. The GD attained the worst performance with the lowest APE values of 98.25, 95.45, 96.10, and 95.70 in the tested locations. In terms of objective function values and their evaluation count, the QN algorithm shows the fewest function counts of 44, 44, 56, and 44, and the lowest objective values of 80.85, 37.77, 54.69, and 41.24, thus attaining the best optimization algorithm results across the study locations. The worst performance was attained by the GD with objective values of 86.45, 39.58, 76.66, and 54.27, respectively. Though the objective values achieved with global optimization methods, PAS, GA, PS, and SA, are relatively small compared to the QN, their function evaluation counts are high. The PAS, GA, PS, and SA recorded 1367, 2550, 3450, and 2818 function evaluation counts, which are relatively high. Overall, the QN algorithm achieves the best optimization, and it can serve as a reference for RF engineers in selecting suitable optimization methods for propagation modeling and parametric tuning.

Meta-evaluation for 3D Face Reconstruction Via Synthetic Data.

The standard benchmark metric for 3D face reconstruction is the geometric error between reconstructed meshes and the ground truth. Nearly all recent reconstruction methods are validated on real ground truth scans, in which case one needs to establish point correspondence prior to error computation, which is typically done with the Chamfer (i.e., nearest neighbor) criterion. However, a simple yet fundamental question have not been asked: Is the Chamfer error an appropriate and fair benchmark metric for 3D face reconstruction? More generally, how can we determine which error estimator is a better benchmark metric? We present a meta-evaluation framework that uses synthetic data to evaluate the quality of a geometric error estimator as a benchmark metric for face reconstruction. Further, we use this framework to experimentally compare four geometric error estimators. Results show that the standard approach not only severely underestimates the error, but also does so inconsistently across reconstruction methods, to the point of even altering the ranking of the compared methods. Moreover, although non-rigid ICP leads to a metric with smaller estimation bias, it could still not correctly rank all compared reconstruction methods, and is significantly more time consuming than Chamfer. In sum, we show several issues present in the current benchmarking and propose a procedure using synthetic data to address these issues.

RETRACTED ARTICLE: Learnings from Industry 4.0 for transitioning towards Industry 4.0+: challenges and solutions for Indian pharmaceutical sector

Industry 4.0 (I4.0) is helping production units to become smarter using cyber-physical systems and cognitive intelligence. The advanced diagnostics with I4.0 technologies (I4.0t) help in making the process highly flexible, resilient and autonomous. Still, the adoption of I4.0t especially in emerging economies like India is at a very slow pace. The present research has used an integrated approach i.e., Analytical Hierarchy Process-Combinative Distance-Based Assessment-Decision-Making Trial and Evaluation Laboratory to propose a barrier solution framework using data from pharmaceutical manufacturing sector. The findings reveal that "Costly venture" is found to be the most critical deterrent while "Customer awareness and satisfaction" is one of the potential solutions for I4.0t adoption. Further, lack of standardisation and fair benchmarking policies especially in the context of developing economies needs immediate attention. This article concludes by proposing a framework which will help to move from I4.0 towards Industry 4.0 + (I4.0+) which emphasises on role of collaboration between man and machine. And leads to sustainable supply chain management.

TinderAI: Support System for Matching AI Algorithms and Embedded Devices


 
 
 Artificial Intelligence (AI) is becoming increasingly important and pervasive in the modern world. The widespread adoption of AI algorithms is reflected in the extensive range of HW devices on which they can be deployed, from high-performance computing nodes to low-power embedded devices. Given the large set of heterogeneous resources where AI algorithms can be deployed, finding the most suitable device and its con- figuration is challenging, even for experts.
 We propose a data-driven approach to assist AI adopters and developers in choosing the optimal HW resource. Our approach is based on three key elements: i) fair benchmarking of target AI algorithms on a set of hetero- geneous platforms, ii) creation of ML models to learn the behaviour of these AI algorithms, and iii) support guidelines to help identify the best deployment option for a given AI algorithm. We demonstrate our approach on a specific (and relevant) use case: Deep Neural Net- work (DNN) inference on embedded devices.
 
 

ForeTiS: A comprehensive time series forecasting framework in Python

Summary: Time series forecasting is a research area with applications in various domains, nevertheless without yielding a predominant method so far. We present ForeTiS, a comprehensive and open source Python framework that allows rigorous training, comparison, and analysis of state-of-the-art time series forecasting approaches. Our framework includes fully automated yet configurable data preprocessing and feature engineering. In addition, we use advanced Bayesian optimization for automatic hyperparameter search. ForeTiS is easy to use, even for non-programmers, requiring only a single line of code to apply state-of-the-art time series forecasting. Various prediction models, ranging from classical forecasting approaches to machine learning techniques and deep learning architectures, are already integrated. More importantly, as a key benefit for researchers aiming to develop new forecasting models, ForeTiS is designed to allow for rapid integration and fair benchmarking in a reliable framework. Thus, we provide a powerful framework for both end users and forecasting experts.Availability:ForeTiS is available at https://github.com/grimmlab/ForeTiS. We provide a setup using Docker, as well as a Python package at https://pypi.org/project/ForeTiS/. Extensive online documentation with hands-on tutorials and videos can be found at https://foretis.readthedocs.io.

A More General Quantum Credit Risk Analysis Framework

Credit risk analysis (CRA) quantum algorithms aim at providing a quadratic speedup over classical analogous methods. Despite this, experts in the business domain have identified significant limitations in the existing approaches. Thus, we proposed a new variant of the CRA quantum algorithm to address these limitations. In particular, we improved the risk model for each asset in a portfolio by enabling it to consider multiple systemic risk factors, resulting in a more realistic and complex model for each asset's default probability. Additionally, we increased the flexibility of the loss-given-default input by removing the constraint of using only integer values, enabling the use of real data from the financial sector to establish fair benchmarking protocols. Furthermore, all proposed enhancements were tested both through classical simulation of quantum hardware and, for this new version of our work, also using QPUs from IBM Quantum Experience in order to provide a baseline for future research. Our proposed variant of the CRA quantum algorithm addresses the significant limitations of the current approach and highlights an increased cost in terms of circuit depth and width. In addition, it provides a path to a substantially more realistic software solution. Indeed, as quantum technology progresses, the proposed improvements will enable meaningful scales and useful results for the financial sector.

The comparison and use of tools for quantification of antimicrobial use in Indonesian broiler farms.

Indonesia has a large broiler industry with extensive antimicrobial use (AMU) according to empirical evidence. However, there are no quantitative data of on-farm AMU. Quantification of AMU at farm level is crucial to guide interventions on antimicrobial stewardship (AMS). The objective of this study was to compare on-farm AMU monitoring methods, to assess which monitoring method is best suited to gain insight in the quantitative AMU at farm level in medium-scale Indonesian broiler farms. AMU was calculated using four different indicators-mg/PCU (mass-based), TFUDDindo (Treatment Frequency of Used Daily Dose, dose-based), TFDDDvet (Treatment Frequency of Defined Daily Dose, dose-based), and TFcount - based (count-based)-for the total AMU of 98 production cycles with an average length of 30 days. Broilers were exposed to an average of 10 days of antimicrobial treatments per production cycle, whereas 60.8% of the antimicrobials belonged to the Highest Priority Critically Important Antimicrobials (HPCIAs). For each pair of indicators, the Spearman rank correlation coefficient was calculated to assess if the production cycles were ranked consistently in increasing AMU across the different indicators. The correlation varied between 0.4 and 0.8. This study illustrates the considerable difference in the ranking of AMU between the different indicators. In a setting comparable to medium-scale broiler farms in Indonesia, where resources are scarce and there is no professional oversight, the TFcount - based method is best suitable. Before implementing an AMU monitoring method, careful consideration of the use-indicators is paramount to achieve fair benchmarking.

Learn2Reg: comprehensive multi-task medical image registration challenge, dataset and evaluation in the era of deep learning.

Image registration is a fundamental medical image analysis task, and a wide variety of approaches have been proposed. However, only a few studies have comprehensively compared medical image registration approaches on a wide range of clinically relevant tasks. This limits the development of registration methods, the adoption of research advances into practice, and a fair benchmark across competing approaches. The Learn2Reg challenge addresses these limitations by providing a multi-task medical image registration data set for comprehensive characterisation of deformable registration algorithms. A continuous evaluation will be possible at https:// learn2reg.grand-challenge.org. Learn2Reg covers a wide range of anatomies (brain, abdomen, and thorax), modalities (ultrasound, CT, MR), availability of annotations, as well as intra- and inter-patient registration evaluation. We established an easily accessible framework for training and validation of 3D registration methods, which enabled the compilation of results of over 65 individual method submissions from more than 20 unique teams. We used a complementary set of metrics, including robustness, accuracy, plausibility, and runtime, enabling unique insight into the current state-of-the-art of medical image registration. This paper describes datasets, tasks, evaluation methods and results of the challenge, as well as results of further analysis of transferability to new datasets, the importance of label supervision, and resulting bias. While no single approach worked best across all tasks, many methodological aspects could be identified that push the performance of medical image registration to new state-of-the-art performance. Furthermore, we demystified the common belief that conventional registration methods have to be much slower than deep-learning-based methods.

Is There Value in Large Cap Index Selection?

The rise of passive investing is shining a new light on the world of Indexing. Indexing is all around us today, whether being discussed by a media outlet showing the movements of the day or a newly launched passive ETF that tracks a recently created index. But there is little research on whether picking one index over another provides value to the investor when the indexes all share very similar index construction methodologies. Is there value in selecting one over the other or should investors choose the one that is the most accessible or cheapest to access? This article aims to fill that gap by conducting a comparative analysis of US Large Cap indexes that are substantially similar in their construction methodologies. As expected, the indexes selected perform very similar to one another based on the quantitative methods used within the analysis. Other than number of securities, the largest difference among the indexes is their exposure to Mid-Cap stocks. The decision of which index to select is dependent on the application being selected. This article focuses on the four purposes outlined by DiBartolomeo (2019) which are: journalistic, representation of an asset class, a fair benchmark, and basis of passive investment. In most cases, each index is a reasonable substitution for the others. Given that risk and return profiles are similar, factors such as number of securities, licensing costs, and brand value are the most significant to consider. This is especially true when selecting an index as the basis of a passive product, where those factors affect the product cost and capacity that can determine the success or failure of a passive product.

The Curriculum–TIMSS Alignment for Grade 8 Science from 2007 to 2019: A Case of South Africa and Singapore

The present study explores the validity of inferences that may be drawn from the Trends in International Mathematics and Science Study (TIMSS) learner achievement data by measuring the South African curriculum–TIMSS alignment for the Grade 8 science assessments from 2007 to 2019. The Singapore curriculum–TIMSS alignment indices are a fair benchmark for comparison, as Singapore is the best-performing country in the TIMSS. The research questions that guide the present study included the calculation and comparison of the South African curriculum–TIMSS and Singapore curriculum–TIMSS alignment indices. The study analysed 585 South African curriculum items, 659 Singapore curriculum items and 1025 TIMSS assessment items from 2007 to 2019. The present study employed four raters and achieved an inter-rater reliability coefficient of 0.85. The South African curriculum–TIMSS analysis achieved an average content alignment index of 0.34, and an average cognitive alignment index of 0.54. As a benchmark, the Singapore curriculum–TIMSS analysis achieved an average content alignment index of 0.36 and an average cognitive alignment index of 0.74. The literature reviewed supports acceptable content and cognitive alignment indices of 0.67 and 0.65, respectively. The present study concludes that valid inferences from learner performance in the TIMSS may not be drawn for the case of South Africa, and even for Singapore valid inferences are based on cognitive alignment only. However, South Africa’s continued participation in the TIMSS is encouraged as the literature supports many benefits associated with the TIMSS non-achievement data.

Evaluation of advanced control strategies for building energy systems

Advanced building control strategies like model predictive control and reinforcement learning can consider forecasts for weather, occupancy, and energy prices. Combined with system and domain knowledge, this makes them a promising approach to reduce buildings’ energy consumption and CO2 emissions. For this reason, model predictive control and reinforcement learning have recently gained more popularity in the scientific literature. Nevertheless, publications often lack comparability among different control algorithms. The studies in the literature mainly focus on the comparison of an advanced algorithm with a conventional alternative. At the same time, use cases and key performance indicators vary strongly. This paper extensively evaluates six advanced control algorithms based on quantitative and qualitative key performance indicators. The considered control algorithms are a state-of-the-art model-free reinforcement learning algorithm (Soft-Actor-Critic), three model predictive controllers based on white-box, gray-box, and black-box modeling, approximate model predictive control, and a well-designed rule-based controller for fair benchmarking. The controllers are applied to an exemplary multi–input–multi–output building energy system and evaluated using a one-year simulation to cover seasonal effects. The considered building energy system is an office room supplied with heat and cold by an air handling unit and a concrete core activation.We consider the violation of air temperature constraints as thermal discomfort, the yearly energy consumption, and the computational effort as quantitative key performance indicators. Compared to the well-tuned rule-based controller, all advanced controllers decrease thermal discomfort. The black-box model predictive controller achieves the highest energy savings with 8.4%, followed by the white-box model predictive controller with 7.4% and the gray-box controller with 7.2%. The reinforcement learning algorithm reduces energy consumption by 7.1% and the approximate model predictive controller by 4.8%. Next to these quantitative key performance indicators, we introduce qualitative criteria like adaptability, interpretability, and required know-how. Furthermore, we discuss the shortcomings and potential improvements of each controller.

Towards practical Quantum Credit Risk Analysis

In recent years a CRA (Credit Risk Analysis) quantum algorithm with a quadratic speedup over classical analogous methods has been introduced [1]. We propose a new variant of this quantum algorithm with the intent of overcoming some of the most significant limitations (according to business domain experts) of this approach. In particular, we describe a method to implement a more realistic and complex risk model for the default probability of each portfolio’s asset, capable of taking into account multiple systemic risk factors. In addition, we present a solution to increase the flexibility of one of the model’s inputs, the Loss Given Default, removing the constraint to use integer values. This specific improvement addresses the need to use real data coming from the financial sector in order to establish fair benchmarking protocols.Although these enhancements come at a cost in terms of circuit depth and width, they nevertheless show a path towards a more realistic software solution. Recent progress in quantum technology shows that eventually, the increase in the number and reliability of qubits will allow for useful results and meaningful scales for the financial sector, also on real quantum hardware, paving the way for a concrete quantum advantage in the field.The paper also describes experiments conducted on simulators to test the circuit proposed and contains an assessment of the scalability of the approach presented.

Fair Benchmark for Unsupervised Node Representation Learning

Most machine-learning algorithms assume that instances are independent of each other. This does not hold for networked data. Node representation learning (NRL) aims to learn low-dimensional vectors to represent nodes in a network, such that all actionable patterns in topological structures and side information can be preserved. The widespread availability of networked data, e.g., social media, biological networks, and traffic networks, along with plentiful applications, facilitate the development of NRL. However, it has become challenging for researchers and practitioners to track the state-of-the-art NRL algorithms, given that they were evaluated using different experimental settings and datasets. To this end, in this paper, we focus on unsupervised NRL and propose a fair and comprehensive evaluation framework to systematically evaluate state-of-the-art unsupervised NRL algorithms. We comprehensively evaluate each algorithm by applying it to three evaluation tasks, i.e., classification fine tuned via a validation set, link prediction fine-tuned in the first run, and classification fine tuned via link prediction. In each task and each dataset, all NRL algorithms were fine-tuned using a random search within a fixed amount of time. Based on the results for three tasks and eight datasets, we evaluate and rank thirteen unsupervised NRL algorithms.

Benchmarking and Interpreting End-to-End Learning of MIMO and Multi-User Communication

End-to-end autoencoder (AE) learning has the potential of exceeding the performance of human-engineered transceivers and encoding schemes, without a priori knowledge of communication-theoretic principles. In this work, we aim to understand to what extent and for which scenarios this claim holds true when comparing with fair benchmarks. Our particular focus is on memoryless multiple-input multiple-output (MIMO) and multi-user (MU) systems. Four case studies are considered: two point-to-point (closed-loop and open-loop MIMO) and two MU scenarios (MIMO broadcast and interference channels). For the point-to-point scenarios, we explain some of the performance gains observed in prior work through the selection of improved baseline schemes that include geometric shaping as well as bit and power allocation. For the MIMO broadcast channel, we demonstrate the feasibility of a novel AE method with centralized learning and decentralized execution. Interestingly, the learned scheme performs close to nonlinear vector-perturbation precoding and significantly outperforms conventional zero-forcing. Lastly, we highlight potential pitfalls when interpreting learned communication schemes. In particular, we show that the AE for the considered interference channel learns to avoid interference, albeit in a rotated reference frame. After de-rotating the learned signal constellation of each user, the resulting scheme corresponds to conventional time sharing with geometric shaping.