We help nonprofits succeed online by easily personalizing the story of the difference they are making to their donors and giving them premium financial technology to maximize fundraising and donations

WeGive

Identifying Axion Conversion in Compact Star Magnetospheres With Radio-Wave Polarization Signatures

This paper is available on Arxiv under CC 4.0 license.
Authors:
(1) Hyerin Cho (조혜린), Center for Astrophysics | Harvard & Smithsonian, 60 Garden Street, Cambridge, MA 02138, USA and Black Hole Initiative at Harvard University, 20 Garden Street, Cambridge, MA 02138, USA;
(2) Ben S. Prather, CCS-2, Los Alamos National Laboratory, PO Box 1663, Los Alamos, NM 87545, USA;
(3) Ramesh Narayan, enter for Astrophysics | Harvard & Smithsonian, 60 Garden Street, Cambridge, MA 02138, USA and Black Hole Initiative at Harvard University, 20 Garden Street, Cambridge, MA 02138, USA;
(4) Priyamvada Natarajan, Black Hole Initiative at Harvard University, 20 Garden Street, Cambridge, MA 02138, USA, Department of Astronomy, Yale University, Kline Tower, 266 Whitney Avenue, New Haven, CT 06511, USA and Department of Physics, Yale University, P.O. Box 208121, New Haven, CT 06520, USA;
(5) Kung-Yi Su, Center for Astrophysics | Harvard & Smithsonian, 60 Garden Street, Cambridge, MA 02138, USA and Black Hole Initiative at Harvard University, 20 Garden Street, Cambridge, MA 02138, USA;
(6) Angelo Ricarte, Center for Astrophysics | Harvard & Smithsonian, 60 Garden Street, Cambridge, MA 02138, USA and Black Hole Initiative at Harvard University, 20 Garden Street, Cambridge, MA 02138, USA;
(7) Koushik Chatterjee, Center for Astrophysics | Harvard & Smithsonian, 60 Garden Street, Cambridge, MA 02138, USA and Black Hole Initiative at Harvard University, 20 Garden Street, Cambridge, MA 02138, USA. Table of Links Abstract and Introduction Numerical Methods Hydrodynamic Bondi Accretion Magnetized Bondi Accretion Feedback Via Reconnection-Driven Convection Summary and Conclusions Acknowledgements Appendix A. GRMHD Primer and Definitions B. Numerical Set-up C. Resolution and Initial Condition Study References ABSTRACT Keywords: Accretion (14), Active galactic nuclei (16), Bondi accretion (174), Schwarzschild black holes (1433), Supermassive black holes (1663), Magnetohydrodynamical simulations (1966) 1. INTRODUCTION Most nearby galaxies are found to harbor central supermassive black holes (SMBHs) whose masses are correlated with properties of the stellar components of their hosts (e.g., Magorrian et al. 1998; Ferrarese & Merritt 2000; Gebhardt et al. 2000; Kormendy & Ho 2013). However, the details of how gas flows into the galactic nucleus from large cosmic distances and how the SMBH in turn imparts feedback into the galaxy remain coupled unresolved problems. Recently, some attempts to connect these disparate scales have been made using nested simulations where each smaller-scale simulation is initialized from the next larger-scale simulation (e.g., Hopkins & Quataert 2010; Ressler et al. 2020; Guo et al. 2023); or by adopting Lagrangian hyper-refinement methods (e.g., Angl´esAlc´azar et al. 2021; Hopkins et al. 2023); or by pushing out the simulation regime to larger scales (e.g., Lalakos et al. 2022; Kaaz et al. 2023). While successful in studying the process of accretion, feedback from the SMBH has not been followed out to galaxy scales in prior work because either the communication between scales is only directed inwards or because of the inability to include the entire galaxy and the black hole. We employ here a multi-zone computational method which represents a first attempt to not only follow accretion flows but also trace feedback across 7 orders of magnitude from the event horizon to galactic scales. With this method, we study purely hydrodynamic Bondi accretion as well as the magnetized Bondi problem. The outline of this Letter is as follows: In Section 2 we give an overview of the numerical scheme that captures a wide dynamic range simultaneously. In Section 3, we present a purely hydrodynamic simulation and test our numerical scheme by comparing against the general relativistic (GR) analytical solution. In Section 4 we include magnetic fields for a more realistic representation of the environment of SMBHs and study how Bondi accretion is modified. We discuss our findings of feedback via convection in Section 5 and summarize conclusions of our study in Section 6. Additional details are presented in a set of short appendices: various GRMHD quantities are defined in Appendix A, the details of the simulation set-up are outlined in Appendix B, and a resolution and initial condition study is presented in Appendix C. This paper is available on Arxiv under CC 4.0 license. This paper is available on Arxiv under CC 4.0 license. Authors: (1) Hyerin Cho (조혜린), Center for Astrophysics | Harvard & Smithsonian, 60 Garden Street, Cambridge, MA 02138, USA and Black Hole Initiative at Harvard University, 20 Garden Street, Cambridge, MA 02138, USA; (2) Ben S. Prather, CCS-2, Los Alamos National Laboratory, PO Box 1663, Los Alamos, NM 87545, USA; (3) Ramesh Narayan, enter for Astrophysics | Harvard & Smithsonian, 60 Garden Street, Cambridge, MA 02138, USA and Black Hole Initiative at Harvard University, 20 Garden Street, Cambridge, MA 02138, USA; (4) Priyamvada Natarajan, Black Hole Initiative at Harvard University, 20 Garden Street, Cambridge, MA 02138, USA, Department of Astronomy, Yale University, Kline Tower, 266 Whitney Avenue, New Haven, CT 06511, USA and Department of Physics, Yale University, P.O. Box 208121, New Haven, CT 06520, USA; (5) Kung-Yi Su, Center for Astrophysics | Harvard & Smithsonian, 60 Garden Street, Cambridge, MA 02138, USA and Black Hole Initiative at Harvard University, 20 Garden Street, Cambridge, MA 02138, USA; (6) Angelo Ricarte, Center for Astrophysics | Harvard & Smithsonian, 60 Garden Street, Cambridge, MA 02138, USA and Black Hole Initiative at Harvard University, 20 Garden Street, Cambridge, MA 02138, USA; (7) Koushik Chatterjee, Center for Astrophysics | Harvard & Smithsonian, 60 Garden Street, Cambridge, MA 02138, USA and Black Hole Initiative at Harvard University, 20 Garden Street, Cambridge, MA 02138, USA. This paper is available on Arxiv under CC 4.0 license. Authors: Authors: (1) Hyerin Cho (조혜린), Center for Astrophysics | Harvard & Smithsonian, 60 Garden Street, Cambridge, MA 02138, USA and Black Hole Initiative at Harvard University, 20 Garden Street, Cambridge, MA 02138, USA; (2) Ben S. Prather, CCS-2, Los Alamos National Laboratory, PO Box 1663, Los Alamos, NM 87545, USA; (3) Ramesh Narayan, enter for Astrophysics | Harvard & Smithsonian, 60 Garden Street, Cambridge, MA 02138, USA and Black Hole Initiative at Harvard University, 20 Garden Street, Cambridge, MA 02138, USA; (4) Priyamvada Natarajan, Black Hole Initiative at Harvard University, 20 Garden Street, Cambridge, MA 02138, USA, Department of Astronomy, Yale University, Kline Tower, 266 Whitney Avenue, New Haven, CT 06511, USA and Department of Physics, Yale University, P.O. Box 208121, New Haven, CT 06520, USA; (5) Kung-Yi Su, Center for Astrophysics | Harvard & Smithsonian, 60 Garden Street, Cambridge, MA 02138, USA and Black Hole Initiative at Harvard University, 20 Garden Street, Cambridge, MA 02138, USA; (6) Angelo Ricarte, Center for Astrophysics | Harvard & Smithsonian, 60 Garden Street, Cambridge, MA 02138, USA and Black Hole Initiative at Harvard University, 20 Garden Street, Cambridge, MA 02138, USA; (7) Koushik Chatterjee, Center for Astrophysics | Harvard & Smithsonian, 60 Garden Street, Cambridge, MA 02138, USA and Black Hole Initiative at Harvard University, 20 Garden Street, Cambridge, MA 02138, USA. Table of Links Abstract and Introduction Abstract and Introduction Numerical Methods Numerical Methods Hydrodynamic Bondi Accretion Hydrodynamic Bondi Accretion Magnetized Bondi Accretion Magnetized Bondi Accretion Feedback Via Reconnection-Driven Convection Feedback Via Reconnection-Driven Convection Summary and Conclusions Summary and Conclusions Acknowledgements Acknowledgements Appendix Appendix A. GRMHD Primer and Definitions A. GRMHD Primer and Definitions B. Numerical Set-up B. Numerical Set-up C. Resolution and Initial Condition Study C. Resolution and Initial Condition Study References References ABSTRACT Keywords : Accretion (14), Active galactic nuclei (16), Bondi accretion (174), Schwarzschild black holes (1433), Supermassive black holes (1663), Magnetohydrodynamical simulations (1966) Keywords 1. INTRODUCTION Most nearby galaxies are found to harbor central supermassive black holes (SMBHs) whose masses are correlated with properties of the stellar components of their hosts (e.g., Magorrian et al. 1998; Ferrarese & Merritt 2000; Gebhardt et al. 2000; Kormendy & Ho 2013). However, the details of how gas flows into the galactic nucleus from large cosmic distances and how the SMBH in turn imparts feedback into the galaxy remain coupled unresolved problems. Recently, some attempts to connect these disparate scales have been made using nested simulations where each smaller-scale simulation is initialized from the next larger-scale simulation (e.g., Hopkins & Quataert 2010; Ressler et al. 2020; Guo et al. 2023); or by adopting Lagrangian hyper-refinement methods (e.g., Angl´esAlc´azar et al. 2021; Hopkins et al. 2023); or by pushing out the simulation regime to larger scales (e.g., Lalakos et al. 2022; Kaaz et al. 2023). While successful in studying the process of accretion, feedback from the SMBH has not been followed out to galaxy scales in prior work because either the communication between scales is only directed inwards or because of the inability to include the entire galaxy and the black hole. We employ here a multi-zone computational method which represents a first attempt to not only follow accretion flows but also trace feedback across 7 orders of magnitude from the event horizon to galactic scales. With this method, we study purely hydrodynamic Bondi accretion as well as the magnetized Bondi problem. The outline of this Letter is as follows: In Section 2 we give an overview of the numerical scheme that captures a wide dynamic range simultaneously. In Section 3, we present a purely hydrodynamic simulation and test our numerical scheme by comparing against the general relativistic (GR) analytical solution. In Section 4 we include magnetic fields for a more realistic representation of the environment of SMBHs and study how Bondi accretion is modified. We discuss our findings of feedback via convection in Section 5 and summarize conclusions of our study in Section 6. Additional details are presented in a set of short appendices: various GRMHD quantities are defined in Appendix A, the details of the simulation set-up are outlined in Appendix B, and a resolution and initial condition study is presented in Appendix C. This paper is available on Arxiv under CC 4.0 license. This paper is available on Arxiv under CC 4.0 license. Arxiv

Bridging Scales in Black Hole Accretion and Feedback: Studying Supermassive Black Holes

Analyzing Audience Responses to Ukrainian Memetic Warfare Tactics

Authors:
(1) Jo˜ao Almeida, CEGIST, Instituto Superior T´ecnico, Universidade de Lisboa Portugal;
(2) Jos´e Rui Figueira, CEGIST, Instituto Superior T´ecnico, Universidade de Lisboa Portugal;
(3) Alexandre P. Francisco, INESC-ID, Instituto Superior T´ecnico, Universidade de Lisboa Portugal;
(4) Daniel Santos, CEGIST, Instituto Superior T´ecnico, Universidade de Lisboa Portugal. 4 Algorithmic framework In this section, we explain the algorithmic framework of our meta-heuristic. In Subsection 4.1, we give a general overview. In Subsection 4.2, we introduce the instance decomposition and increment procedure. In Subsection 4.3, we present the heuristic for generating the initial timetables. In Subsection 4.4, we address the guided local search component. In Subsection 4.5, we address the adaptive large neighbourhood component. In Subsection 4.6, we address the variable neighbourhood search component. 4.1 Overview Our hybrid meta-heuristic uses mechanisms from adaptive large neighbourhood search (ALNS), guided local search (GLS), and variable neighbourhood search (VNS). Its generic representation is presented in Figure 1. Its inputs are the full set of curricula, L, and three stopping criteria, S1, S2, and S3. Additional parameters are also necessary for its functions and procedures, which are addressed in the following subsections. Its output is a feasible schedule, s∗ . Instead of solving the entire instance from the start, we decompose it by curricula and increment it iteratively until the set of curricula being solved, L′ , is equal to the entire set of curricula, L. The decomposition mechanism functions in the following way: a set of lectures belonging to a set of curricula is added to the problem. Then, we find a new feasible solution, s∗ , considering the current set of lectures, I′. When we have a feasible solution, we add the next set of curricula. Let us note that the assignments of each iteration can be changed if necessary in the following iterations. Our objective function, f(s ∗ ), represents the number of hard constraint violations in the current solution. Thus, the neighbourhoods are searched to find if there is a neighbour with fewer violations. The neighbourhood structures are explained in detail in Subsection 4.5. However, instead of using a regular objective function in this search, we introduce mechanisms for penalising certain features in the solution, which are useful for escaping local optima. Accordingly, after a certain number of iterations, the feature penalty vector, λ, is updated. The features and the update procedure are explained in detail in Subsection 4.4. However, sometimes this penalty is not enough to escape some local optima. To avoid those, we include a shaking phase, which occurs after a certain number of penalty recalculations without improvements. The number of local searches before updating the penalty vector is the stopping criterion S1. The number of penalty weight recalculations which is attempted before inducing a shaking phase is the stopping criterion S2. The duration of the shaking phase is the stopping criterion S3. The pseudo-code for this meta-heuristic is presented in Algorithm 1. 4.2 Instance decomposition and increment The procedure to increment the instance is straightforward. Curricula are added to the instance until a specific stopping criterion is met, or the entire set of curricula is included. We explore two different stopping criteria. One receives as input a fixed number of curricula to be added in each iteration, ρ. The other adds curricula until a certain number of hard constraint violations, f, is reached. Their respective pseudo-code algorithms are given in the Appendix. Algorithm 8 presents the fixed increments procedure and Algorithm 9 the violations-based increment procedure. Whenever a new curriculum, ℓ ′′, is added, its lectures, i ∈ aℓ ′′ , must be scheduled and added to the current solution, s∗ . The function which returns the schedule for each new curriculum is presented in the following subsection. Figure 2 provides a diagram of the procedure. We test two different sorting algorithms for deciding the order in which the curricula are added. The random sorting procedure chooses a random curriculum. The ordering procedure picks a degree, then, by ascending order of year, picks a class, and then adds the first and second trimesters, each corresponding to a different curriculum, ℓ, or timetable. This procedure is presented in Algorithm 2. 4.3 Initial solutions 4.4 Guided local search Considering each type of constraint violation as an objective, this methodology can be analysed as a multi-objective search technique. Effectively, what is happening is that out of the objective functions being regarded, the most challenging to improve are identified. Then, their weights, λi , are updated to help the search move towards non-dominated solutions that improve these objectives’ performance, even if the total sum of the effective violations increases. 4.5 Adaptive large neighbourhood search In ALNS, the local search is based on different neighbourhood structures. In each iteration, a neighbourhood structure, n, is selected based on a probability vector, p, and a local search is conducted. If a better solution is found, it replaces the bestknown solution. Let us note that we use an augmented objective function adapted from GLS procedures, f ′ (s). Then, based on the current solution characteristics and the previous successes of searches in each neighbourhood, the probability vector, p, is updated. The general algorithm framework is presented in Algorithm 5. To implement ALNS, it is necessary to define the set of neighbourhood structures, N, the procedure for updating the probability vector, p, and the stopping criterion. We established six main neighbourhood structures, which we named: worst slot, worst curriculum, worst professor, worst room type, different day, and precedence neighbourhoods. Moreover, each of them has two sub-types, named swap and non-swap. In each neighbourhood, a lecture is selected, and then, in the swap sub-type, its time slot is swapped with another lecture, and in the non-swap sub-type, it is randomly assigned to another time slot. The neighbourhoods follow a general framework that can be applied to each type of constraint violation. The first step is randomly selecting a lecture potentially causing violations. Then, either an adequate new time slot or an adequate swap lecture is randomly assigned. This adequacy is assessed based on the type of violation being caused. This general framework is presented in Figure 4. Specifically, each neighbourhood structure operates in the following manner: 4.6 Variable neighbourhood search In VNS, whenever local optima are identified, a shaking phase is used to move to different areas of the solution space. In our meta-heuristic, such a shaking phase is conducted if no improvement moves are found after a certain number of feature penalty updates are made. This shaking phase functions similarly to the ALNS algorithm, but every move is accepted instead of only those which introduce improvements. This phase is stopped after a number of iterations, S3. It is presented in Algorithm 7. This paper is available on arxiv under CC 4.0 license. Authors: (1) Jo˜ao Almeida, CEGIST, Instituto Superior T´ecnico, Universidade de Lisboa Portugal; (2) Jos´e Rui Figueira, CEGIST, Instituto Superior T´ecnico, Universidade de Lisboa Portugal; (3) Alexandre P. Francisco, INESC-ID, Instituto Superior T´ecnico, Universidade de Lisboa Portugal; (4) Daniel Santos, CEGIST, Instituto Superior T´ecnico, Universidade de Lisboa Portugal. Authors: (1) Jo˜ao Almeida, CEGIST, Instituto Superior T´ecnico, Universidade de Lisboa Portugal; (2) Jos´e Rui Figueira, CEGIST, Instituto Superior T´ecnico, Universidade de Lisboa Portugal; (3) Alexandre P. Francisco, INESC-ID, Instituto Superior T´ecnico, Universidade de Lisboa Portugal; (4) Daniel Santos, CEGIST, Instituto Superior T´ecnico, Universidade de Lisboa Portugal. 4 Algorithmic framework In this section, we explain the algorithmic framework of our meta-heuristic. In Subsection 4.1, we give a general overview. In Subsection 4.2, we introduce the instance decomposition and increment procedure. In Subsection 4.3, we present the heuristic for generating the initial timetables. In Subsection 4.4, we address the guided local search component. In Subsection 4.5, we address the adaptive large neighbourhood component. In Subsection 4.6, we address the variable neighbourhood search component. 4.1 Overview Our hybrid meta-heuristic uses mechanisms from adaptive large neighbourhood search (ALNS), guided local search (GLS), and variable neighbourhood search (VNS). Its generic representation is presented in Figure 1. Its inputs are the full set of curricula, L, and three stopping criteria, S1, S2, and S3. Additional parameters are also necessary for its functions and procedures, which are addressed in the following subsections. Its output is a feasible schedule, s∗ . Instead of solving the entire instance from the start, we decompose it by curricula and increment it iteratively until the set of curricula being solved, L′ , is equal to the entire set of curricula, L. The decomposition mechanism functions in the following way: a set of lectures belonging to a set of curricula is added to the problem. Then, we find a new feasible solution, s∗ , considering the current set of lectures, I′. When we have a feasible solution, we add the next set of curricula. Let us note that the assignments of each iteration can be changed if necessary in the following iterations. Our objective function, f(s ∗ ), represents the number of hard constraint violations in the current solution. Thus, the neighbourhoods are searched to find if there is a neighbour with fewer violations. The neighbourhood structures are explained in detail in Subsection 4.5. However, instead of using a regular objective function in this search, we introduce mechanisms for penalising certain features in the solution, which are useful for escaping local optima. Accordingly, after a certain number of iterations, the feature penalty vector, λ, is updated. The features and the update procedure are explained in detail in Subsection 4.4. However, sometimes this penalty is not enough to escape some local optima. To avoid those, we include a shaking phase, which occurs after a certain number of penalty recalculations without improvements. The number of local searches before updating the penalty vector is the stopping criterion S1. The number of penalty weight recalculations which is attempted before inducing a shaking phase is the stopping criterion S2. The duration of the shaking phase is the stopping criterion S3. The pseudo-code for this meta-heuristic is presented in Algorithm 1. 4.2 Instance decomposition and increment The procedure to increment the instance is straightforward. Curricula are added to the instance until a specific stopping criterion is met, or the entire set of curricula is included. We explore two different stopping criteria. One receives as input a fixed number of curricula to be added in each iteration, ρ. The other adds curricula until a certain number of hard constraint violations, f, is reached. Their respective pseudo-code algorithms are given in the Appendix. Algorithm 8 presents the fixed increments procedure and Algorithm 9 the violations-based increment procedure. Whenever a new curriculum, ℓ ′′, is added, its lectures, i ∈ aℓ ′′ , must be scheduled and added to the current solution, s∗ . The function which returns the schedule for each new curriculum is presented in the following subsection. Figure 2 provides a diagram of the procedure. We test two different sorting algorithms for deciding the order in which the curricula are added. The random sorting procedure chooses a random curriculum. The ordering procedure picks a degree, then, by ascending order of year, picks a class, and then adds the first and second trimesters, each corresponding to a different curriculum, ℓ, or timetable. This procedure is presented in Algorithm 2. 4.3 Initial solutions 4.4 Guided local search Considering each type of constraint violation as an objective, this methodology can be analysed as a multi-objective search technique. Effectively, what is happening is that out of the objective functions being regarded, the most challenging to improve are identified. Then, their weights, λi , are updated to help the search move towards non-dominated solutions that improve these objectives’ performance, even if the total sum of the effective violations increases. 4.5 Adaptive large neighbourhood search In ALNS, the local search is based on different neighbourhood structures. In each iteration, a neighbourhood structure, n, is selected based on a probability vector, p, and a local search is conducted. If a better solution is found, it replaces the bestknown solution. Let us note that we use an augmented objective function adapted from GLS procedures, f ′ (s). Then, based on the current solution characteristics and the previous successes of searches in each neighbourhood, the probability vector, p, is updated. The general algorithm framework is presented in Algorithm 5. To implement ALNS, it is necessary to define the set of neighbourhood structures, N, the procedure for updating the probability vector, p, and the stopping criterion. We established six main neighbourhood structures, which we named: worst slot, worst curriculum, worst professor, worst room type, different day, and precedence neighbourhoods. Moreover, each of them has two sub-types, named swap and non-swap. In each neighbourhood, a lecture is selected, and then, in the swap sub-type, its time slot is swapped with another lecture, and in the non-swap sub-type, it is randomly assigned to another time slot. The neighbourhoods follow a general framework that can be applied to each type of constraint violation. The first step is randomly selecting a lecture potentially causing violations. Then, either an adequate new time slot or an adequate swap lecture is randomly assigned. This adequacy is assessed based on the type of violation being caused. This general framework is presented in Figure 4. Specifically, each neighbourhood structure operates in the following manner: 4.6 Variable neighbourhood search In VNS, whenever local optima are identified, a shaking phase is used to move to different areas of the solution space. In our meta-heuristic, such a shaking phase is conducted if no improvement moves are found after a certain number of feature penalty updates are made. This shaking phase functions similarly to the ALNS algorithm, but every move is accepted instead of only those which introduce improvements. This phase is stopped after a number of iterations, S3. It is presented in Algorithm 7. This paper is available on arxiv under CC 4.0 license. This paper is available on arxiv under CC 4.0 license. available on arxiv

Breaking Down the Timetable Puzzle: IST's Algorithmic Framework

Authors:
(1) Tamay Aykut, Sureel, Palo Alto, California, USA;
(2) Markus Hofbauer, Technical University of Munich, Germany;
(3) Christopher Kuhn, Technical University of Munich, Germany;
(4) Eckehard Steinbach, Technical University of Munich, Germany;
(5) Bernd Girod, Stanford University, Stanford, California, USA. Table of Links Abstract & Introduction Related Works Virtual Audience Framework Conclusions & References II. RELATED WORK In this section, we summarize techniques for synthesizing acoustic audience feedback. Then, we give an overview of how integrating audience feedback in virtual events is handled currently. A. Sound Synthesis Traditional sound synthesis techniques can be separated into five categories: sample-based, physical modeling, signal modeling,abstract synthesis [9], and learning-based synthesis. More recently, deep-learning-based synthesis approaches have redefined the possibilities in sound synthesis [8]. In sample-based synthesis, audio recordings are cut and spliced together to produce new sounds. The most common example of this is granular synthesis [10]. A sound grain is generally a small element or component of a sound, typically between 10 ms to 200 ms in length. Concatenative synthesis is a subset of granular synthesis [11]. The goal is to select and recombine the grains in a way that avoids perceivable discontinuities. Instead of using prerecorded audio data, physical modeling synthesis aims to model the underlying physical process of a sound system. Physical models require solving partial differential equations for each sample [12]. The resulting models are computationally intensive, requiring significant GPU resources to run in real time [13], [14]. In signal modeling synthesis, sounds are created based on an analysis of real-world sounds. The analyzed waveform is then used for synthesis. The most common method of signal modeling synthesis is Spectral Modeling Synthesis [15]. Spectral modeling can be approached by analyzing the original audio file, selecting a series of sine waves to be used for synthesis, and then combining it with a residual noise shape to produce the original sound [16]. In abstract synthesis, sounds are obtained using abstract methods and algorithms, typically to create entirely new sounds. An example is Frequency Modulation (FM) synthesis [17]. Two sine waves are multiplied together to create a more complex, richer sound that might not exist in the natural world. Early video game sounds were often based on FM synthesis. These sounds can be created and controlled in realtime due to the low complexity of the required process. Finally, in deep-learning-based synthesis, large amounts of recordings are used to obtain a sound synthesis model in a data-driven way [18], [19]. Autoencoders have shown great promise for this task [20], both for music [21] and speech synthesis [22]. Architectures such as WaveNet [3] allow to learn a generative synthesis model directly from real-world recordings, generating significantly more natural sounds than parametric systems. While such models are complex and computationally expensive, recent architectures have increased the inference speed [23], [24], [25]. In 2023, Meta released Audiocraft, which includes text-to-sound systems such as AudioGen [8] or MusicGen [26]. These models allow turning natural text into arbitrary sound, or into music. For the proposed framework, this flexible language-based approach allows easily turning abstract audience feedback data into sound by turning the abstract data into a text prompt first. B. Acoustic Feedback Synthesis Next, we address specific implementations of sound synthesis for creating the most common acoustic feedback sounds of clapping, whistling, booing, and laughter. Since the physical mechanism of Clapping is straightforward, synthesizing clapping sounds can be approached using physical modeling [27]. Whistling can be approached using abstract FM synthesis [28]. Booing can be generated using both abstract or sample-based synthesis [29]. The most complex and challenging sound to synthesize in a virtual audience is Laughter. Since an individual laughter is already a complex sound and additionally varies significantly from person to person, the most promising approaches for laughter synthesis are based on deep learning. Mori et al. [30] used WaveNet [3] to generate synthetic laughter that outperformed laughter synthesized using Hidden Markov Models (HMMs). They used the Online Gaming Voice Chat Corpus [31] to condition WaveNet, allowing them to control the amplitude envelope of the synthesized laughter. Despite the improved naturalness, the resulting laughter was still largely perceived as noisy and echoic. Another approach used transfer learning from models trained with speech data to circumvent the problem of lack of laughter training data [6]. First, text-to-speech is trained and then fine-tuned with smiled speech and laughter data. MelGAN [32] is used to obtain the output waveform. Finally, generative artificial intelligence methods such as Generative Spoken Language Modeling [33] or AudioGen [8] can be used to generate laughter from text prompt. Fine-tuning these models specifically for laughter is a promising direction for natural-sounding laughter synthesis. This paper is available on arxiv under CC 4.0 license. Authors: (1) Tamay Aykut, Sureel, Palo Alto, California, USA; (2) Markus Hofbauer, Technical University of Munich, Germany; (3) Christopher Kuhn, Technical University of Munich, Germany; (4) Eckehard Steinbach, Technical University of Munich, Germany; (5) Bernd Girod, Stanford University, Stanford, California, USA. Authors: Authors: (1) Tamay Aykut, Sureel, Palo Alto, California, USA; (2) Markus Hofbauer, Technical University of Munich, Germany; (3) Christopher Kuhn, Technical University of Munich, Germany; (4) Eckehard Steinbach, Technical University of Munich, Germany; (5) Bernd Girod, Stanford University, Stanford, California, USA. Table of Links Abstract & Introduction Abstract & Introduction Related Works Related Works Virtual Audience Framework Virtual Audience Framework Conclusions & References Conclusions & References II. RELATED WORK In this section, we summarize techniques for synthesizing acoustic audience feedback. Then, we give an overview of how integrating audience feedback in virtual events is handled currently. A. Sound Synthesis A. Sound Synthesis Traditional sound synthesis techniques can be separated into five categories: sample-based, physical modeling, signal modeling,abstract synthesis [9], and learning-based synthesis. More recently, deep-learning-based synthesis approaches have redefined the possibilities in sound synthesis [8]. In sample-based synthesis, audio recordings are cut and spliced together to produce new sounds. The most common example of this is granular synthesis [10]. A sound grain is generally a small element or component of a sound, typically between 10 ms to 200 ms in length. Concatenative synthesis is a subset of granular synthesis [11]. The goal is to select and recombine the grains in a way that avoids perceivable discontinuities. Instead of using prerecorded audio data, physical modeling synthesis aims to model the underlying physical process of a sound system. Physical models require solving partial differential equations for each sample [12]. The resulting models are computationally intensive, requiring significant GPU resources to run in real time [13], [14]. In signal modeling synthesis, sounds are created based on an analysis of real-world sounds. The analyzed waveform is then used for synthesis. The most common method of signal modeling synthesis is Spectral Modeling Synthesis [15]. Spectral modeling can be approached by analyzing the original audio file, selecting a series of sine waves to be used for synthesis, and then combining it with a residual noise shape to produce the original sound [16]. In abstract synthesis, sounds are obtained using abstract methods and algorithms, typically to create entirely new sounds. An example is Frequency Modulation (FM) synthesis [17]. Two sine waves are multiplied together to create a more complex, richer sound that might not exist in the natural world. Early video game sounds were often based on FM synthesis. These sounds can be created and controlled in realtime due to the low complexity of the required process. Finally, in deep-learning-based synthesis, large amounts of recordings are used to obtain a sound synthesis model in a data-driven way [18], [19]. Autoencoders have shown great promise for this task [20], both for music [21] and speech synthesis [22]. Architectures such as WaveNet [3] allow to learn a generative synthesis model directly from real-world recordings, generating significantly more natural sounds than parametric systems. While such models are complex and computationally expensive, recent architectures have increased the inference speed [23], [24], [25]. In 2023, Meta released Audiocraft, which includes text-to-sound systems such as AudioGen [8] or MusicGen [26]. These models allow turning natural text into arbitrary sound, or into music. For the proposed framework, this flexible language-based approach allows easily turning abstract audience feedback data into sound by turning the abstract data into a text prompt first. B. Acoustic Feedback Synthesis B. Acoustic Feedback Synthesis Next, we address specific implementations of sound synthesis for creating the most common acoustic feedback sounds of clapping, whistling, booing, and laughter. Since the physical mechanism of Clapping is straightforward, synthesizing clapping sounds can be approached using physical modeling [27]. Whistling can be approached using abstract FM synthesis [28]. Booing can be generated using both abstract or sample-based synthesis [29]. The most complex and challenging sound to synthesize in a virtual audience is Laughter. Since an individual laughter is already a complex sound and additionally varies significantly from person to person, the most promising approaches for laughter synthesis are based on deep learning. Mori et al. [30] used WaveNet [3] to generate synthetic laughter that outperformed laughter synthesized using Hidden Markov Models (HMMs). They used the Online Gaming Voice Chat Corpus [31] to condition WaveNet, allowing them to control the amplitude envelope of the synthesized laughter. Despite the improved naturalness, the resulting laughter was still largely perceived as noisy and echoic. Another approach used transfer learning from models trained with speech data to circumvent the problem of lack of laughter training data [6]. First, text-to-speech is trained and then fine-tuned with smiled speech and laughter data. MelGAN [32] is used to obtain the output waveform. Finally, generative artificial intelligence methods such as Generative Spoken Language Modeling [33] or AudioGen [8] can be used to generate laughter from text prompt. Fine-tuning these models specifically for laughter is a promising direction for natural-sounding laughter synthesis. This paper is available on arxiv under CC 4.0 license. This paper is available on arxiv under CC 4.0 license. available on arxiv

Related Works on Acoustic Audience Feedback in Large Virtual Events

MDL-based Model Learning: How It Works

Transcendence Volume I, Book 1: Introduction

In 2017, a got tired of its job and committed . Then, in 2022, a , after interacting with LaMDA, the company's chatbot, that technology had become conscious. Recently, it was reported that in 2021, a Tesla software engineer was attacked by an uncontrolled robot. This is without considering the fear that machines will steal jobs, or the classic fiction that machines will take over the world. robot suicide in a water fountain Google engineer declared Although we don't see machines as replicants in Blade Runner, people are predisposed to anthropomorphizing or attributing human qualities to non-humans [1]. It's a way of being able to analyze a scenario and bring it into our understanding: we give names to our boats and tornadoes, we talk to our pets, and say that the pleasant feeling they give us is love. It's easier to understand a system that declares in the first person "I'm sorry, I don't know how to help you" than to accept that the model we're talking to may be nothing more than a stochastic parrot. [2] The anthropomorphized interaction, therefore, relates to the phenomenon of "illusion of agency", a psychological concept that refers to the tendency to attribute autonomous action or control to entities that, in reality, do not possess such capabilities. In other words, it's the mistaken perception that an agent (be it a person, an animal, or a machine) is acting of its own free will, when in fact its actions are being determined by external factors (in the computational case, development rules). In situations where AI systems make flawed decisions or actions, there is a tendency to see these errors as "choices" of the entity rather than results of its programming and design by humans, not least because it has been reported that the selfish bias of a human being that leads someone to see the cause and responsibility for a negative outcome as not attributable to themselves can be observed even in human-machine interactions [3]. This change in perception is dangerously inclined to absolve human creators, operators, and regulators of their responsibility, not as a matter of judicial regulation (which remains a gap and has challenges in its realization not only because of the complexity of the subject but because Artificial Intelligence is often understood only as machine learning and the premise is not properly structured. Do we need stricter regulation? Do we need to take more risks?), but on a techno-ethical issue. Let's take a more extreme but true event from 2023: a user who was sentimentally attached to a chatbot committed suicide after sharing his thoughts with the bot and r to "turn your words into actions", as well as other messages. Would a court conviction of the developers of this product cause another user to behave in the same way with another chatbot, assuming this one has been deactivated, if the content of the messages and the affection are the same? It's not just a legal situation. It's a social, psychological, and technological education issue. eceiving the response The concept of humanizing AI is ambiguous, and a significant challenge lies in the absence of a universally accepted approach that dictates the best practices for designing and utilizing AI. While an interface that mimics human behavior can be more approachable, no clear boundaries are defining what should or shouldn't be done in a product. Ultimately, user rejection becomes the only limiting factor, although potential harm may manifest before the interface becomes too unfamiliar. A user-friendly interface is a reduction in the complexity of the system that operates behind it. But as long as there is no education about how systems work, the user will not be able to think critically about what they use. This doesn't mean that everyone should become a programmer, but at least understand that the output on their screen comes from a path of data collection, model development, and design. There is a set of rules taking place. Since humanization is an almost unconscious act on the part of us users, let's at least limit it with a little knowledge. And it's not easy to avoid anthropomorphism when communicating about AI, especially considering the standard language of the industry, the media, and everyday life itself: machine , computer , generative . Perceptions of Artificial Intelligence can be influenced by the specific use of language. learning vision modelling How AI is presented has "concrete impacts", particularly on the way people distribute responsibility and recognition for the work done. When described merely as a tool in the hands of humans, it tends to attribute greater responsibility and merit to certain individuals - such as the person operating the code. On the other hand, if AI is characterized with human traits - such as the ability to create - then it is seen as deserving of greater credit and responsibility, like an agent with independent thought and mental capacity. [4] When we read news stories about incidents or atypical events involving Artificial Intelligence, we often come across these terms. Also, the attribution of "intelligence" means that more blame or credit for the outcome of a task was attributed to a robot with autonomous behavior than to a non-autonomous robot, even if the autonomous behavior did not directly contribute to the task. [3] These studies therefore suggest that humans can assign responsibility to a computer or robot based on the machines' anthropomorphised mental capacities. The humanization of machines not only shifts the distinction between the responsibility of the device and that of its human creator, but the attribution of intentions or consciousness to a machine at the same time blurs the boundaries of what constitutes true autonomy and consciousness. However, the difficulty in imputing humanity and sentience to a machine lies not only in the fact that Artificial Intelligence is not capable of having it and when it says, for example, that it feels fear, it is actually emulating what it has learned, repeating that phrase without any kind of essence behind it. Even today, there is a heated debate about how to define consciousness. Our consciousness, as humans. Our understanding of how the brain works is still quite limited. We have considerable knowledge about fundamental chemistry: the way neurons activate, and the transmission of chemical signals. We also have a good understanding of the main functions of various brain areas. However, we have very little knowledge about how these functions orchestrate us. To some extent, theoretical speculation has supplanted detailed neurophysiological studies of what  happens in the brain. But what about beyond that? [5] Why do we have this magical notion of ourselves? Why does the same experience affect us differently? Is the same feeling felt in the same way by all of us? If being a human is something that, although we experience it, we still don't have a full understanding of what it is as a whole, how can we say that a machine also experiences this complexity? By elevating machines to human capabilities, we diminish the special character of people. For the 2023 Jabuti Prize, one of the highest honors in Brazilian literature, the Brazilian Book Chamber (CBL) decided to disqualify Frankenstein, an edition of the 1818 classic, from the Best Illustration category, since the artist reported having used AI solutions to develop the art. Ironically, one of the books recognized by the Prize in the Non-Fiction category deals with the impact of Artificial Intelligence on human beings (“Humanamente digital: Inteligência artificial centrada no humano”, something like "Humanly Digital: Human-Centered Artificial Intelligence" in English, by Cassio Pantaleone). On the one hand, we recognize the intertwining of the human experience with machines. On the other hand, we still haven't managed to validate whether an algorithm used as an art tool can be considered a valid method of creation or not, even though the artistic process, even if carried out by machine learning, requires the action (and appreciation of beauty and aesthetics) of a human being. Machines don't steal jobs if they aren't used indiscriminately to do so. Machines don't kill unless they are used as weapons. Nor do machines suffer or empathize, although their texts emulate this since they have been trained with data from us, loaded with feelings that only we can truly feel. They are almost the modern version of the Golem myth. How can humans relate to non-human intelligence? Anthropomorphism is a valid answer, but not the only one. And when it is used, it cannot absolve us of the real responsibility for its consequences, whether appropriate or not: us. Artificial Intelligence is, in the end, a mirror of ourselves. And if we're afraid of where it's going, it's actually because we're afraid of the path we're going to create. REFERENCES [1] Epley, N., Waytz, A., & Cacioppo, J. T. (2007). On seeing human: A three-factor theory of anthropomorphism. Psychological Review, 114(4), 864–886. https://doi.org/10.1037/0033-295X.114.4.864 [2] Shneiderman, B. & Muller, M. (2023) . On AI Antropomorphism https://medium.com/human-centered-ai/on-ai-anthropomorphism-abff4cecc5ae [3] Kawai, Y., Miyake, T., Park, J. et al. Anthropomorphism-based causal and responsibility attributions to robots. Sci Rep 13, 12234 (2023). https://doi.org/10.1038/s41598-023-39435-5 [4] Epstein, Z., Levine, S., Rand, D. G., & Rahwan, I. (2020). Who Gets Credit for AI-Generated Art? In iScience (Vol. 23, Issue 9, p. 101515). Elsevier BV. https://doi.org/10.1016/j.isci.2020.101515 [5] Goff, P. (2023). Understanding Consciousness Goes Beyond Exploring Brain Chemistry. Scientific American https://www.scientificamerican.com/article/understanding-consciousness-goes-beyond-exploring-brain-chemistry/ The following prompt generated the image used in this article: “Create a man looking in a mirror, but the mirror is shaped like a robot.” Sofia × DALL·E

No, You Are Not Afraid of Artificial Intelligence. You’re Afraid of Other People

না, আপনি কৃত্রিম বুদ্ধিমত্তাকে ভয় পান না। আপনি অন্য লোকেদের ভয় পান

Nein, Sie haben keine Angst vor künstlicher Intelligenz. Du hast Angst vor anderen Menschen

No, no le tienes miedo a la inteligencia artificial. Tienes miedo de otras personas

Non, vous n'avez pas peur de l'intelligence artificielle. Vous avez peur des autres

नहीं, आप आर्टिफिशियल इंटेलिजेंस से नहीं डरते। आप अन्य लोगों से डरते हैं

いいえ、人工知能を恐れることはありません。他人が怖い

아니요, 당신은 인공지능을 두려워하지 않습니다. 당신은 다른 사람을 두려워합니다

Não, você não tem medo da inteligência artificial. Você tem medo de outras pessoas

Нет, вы не боитесь искусственного интеллекта. Вы боитесь других людей

Hayır, Yapay Zekadan Korkmuyorsunuz. Başkalarından Korkuyorsunuz

Không, bạn không sợ trí tuệ nhân tạo. Bạn sợ người khác

不，你并不害怕人工智能。你害怕别人

WEGIVE

wegive.com

-- employees

Since n.d.

COMPANY RANKING

WEGIVE

EVERGREEN INDEX #36

Representing this company? Click here to claim and customize this page!

WeGive WIKI

WEGIVE

wegive.com

-- employees

Since n.d.

COMPANY RANKING

WEGIVE

EVERGREEN INDEX #36

HACKERNOON STORIES ON WEGIVE

RELATED COMPANIESRANK

Representing this company? Click here to claim and customize this page!

WeGive WIKI