Systems and Signal Processing

The systems and signal processing portfolio highlighted a wide range of advanced research that is well planned and offers significant likelihood of technical success. While not all projects were at par with top universities and research institutions, specific niche capabilities were found to be world class, for example, ARL is working at the cutting edge in ground-penetrating radar and cognitive EW. There was also quite a bit of excellent work that, even if evolutionary (as most science is) and building on legacy systems, was consistent with the state-of-the-art in the community. Projects demonstrating particular scientific benefit include emerging work to harness sub-terahertz (THz) transmission frequencies, protective EMSS technologies using porous silicon, and fabrication of substantially ruggedized circuits on non-planar substrates. Research plans demonstrate a good understanding of the relevant literature and methods. Some risks are considered regarding the comparably small number of internal ARL systems and signal processing researchers compared to the extramural research activities, as well as potential attrition caused by retirement. The experimentation in ARL’s in-house laboratories is well conceived, and the addition of extramural laboratories and measurements boosts their capability. The laboratory facilities are very good.

Identified opportunities include the following:

• The Air Force Research Laboratory and the Naval Research Laboratory (NRL) both have excellent personnel in systems and signal processing that ARL may want to consider connecting with.
• This focus area may consider working more closely with the network, cyber, and computational sciences (NC&CS) competency to achieve data analytics and networking technologies among different sensor systems.
• ARL intramural researchers could increase their focus on publications and seeking awards as a means of making better connections to the broader research community.
• While the systems and signal processing projects are utilizing artificial intelligence (AI) and machine learning (ML) in the current projects, the AI and ML area is an opportunistic research area for the general EMSS competency, and increased focus on this area would help the general competency objectives. Increasing efforts in AI and ML based EMSS efforts could help projects such as “Cognitive EW Research Roadmap” and “E/H Sensing” (E stands for “electric field” and H stands for “magnetizing field”). Sensing and security capabilities at higher frequencies such as millimeter-wave and terahertz frequencies are also seen as important areas.
• Efforts toward confidence metrics and experimental data collection for training AI/ML efforts are highly critical. Obtaining and employing real measured data is a key discriminator for ARL’s radio frequency machine learning (RFML) research. The strengthening of experimental measurement facilities, both internally and externally in collaboration with extramural partners, toward generating large training data sets for AI/ML algorithm training is suggested. This effort will also be important toward validation of theoretical and computational research being undertaken within ARL’s intramural activities and through extramural efforts.
• The presentations shown during the review did not have a strong data analysis focus and there were no presentations received on manipulation of multi-modal and heterogeneous data sets. A possible areas of research, if not already under way, to address data analysis could be a focus on multi-modality data-fusion approaches enabled by AI/ML, first for RF-spectrum modalities, subsequently for EM-spectrum modalities, and then for non-EM modalities like acoustics, etc. Efforts within some of the projects related to RFML and control algorithms could be increased to focus on incorporation of real-world training and evaluation data, at a minimum to provide experimental validation of theoretical results. While training data sets are difficult and costly to collect, ARL, acting as a repository of that

training data, potentially in partnership with other services, could improve results of the research across its portfolio.

• For the systems and signal processing research thrust, one identified challenge is the division of intramural staff effort across projects, since scientific staff in this focus area is relatively small. Reducing the number of projects that staff are working on may be helpful, as staff can then gain greater expertise in the higher priority scientific areas, which have been identified with concrete contributions to the portfolio’s goals.
• The systems and signal processing research thrust area could be bolstered with additional expertise in EM phenomenology, computational methods, inverse scattering, and generally speaking, quantitative analysis of synthetic aperture radar and ground penetrating radar data. Additionally, AI expertise is critical, as only cursory uses of AI and ML methods were included in the presentation portfolio. Given how much is happening in this area, the absence of work and expertise in that domain was noticeable and ARL may need to consider ways to increase this focus and expertise.
• With the proliferation of many wireless systems and radars, the electromagnetic spectrum has become very congested. Many communication systems and sensor receivers are prone to self-jamming (co-site interference) and jamming by the adversaries. Distributed, ultra-wide-band and multi-band systems with interferer cancelation capabilities and full-duplex operation could be studied for both communication and radar systems. The application and improvement of low-frequency systems for communication in non-line-of-sight and urban scenarios as well as geolocation could also be reexamined.
• One research area that may be helpful is under-utilized sub-millimeter-wave band systems. Such systems can provide very high bandwidth and secure communication as they can produce very narrow beams with relatively small apertures.
• Increasing collaborations with Established Program to Stimulate Competitive Research states may have potential to benefit ARL.

__________________

^a Because hydrogen passivation of the diamond surface is important in order to control the carrier concentration in the channel, it is important to understand the surface chemistry of diamond, especially with respect to hydrogen passivation. It will be important to determine the conditions needed to fully terminate the diamond surface with hydrogen, using techniques such as hydrogen dosing experiments. It will also be important to determine the thermal stability of the hydrogen passivation to understand the stability of the interface during subsequent processing, such as dielectric and contact deposition. The thermal stability of the passivation can be determined using techniques such as thermal desorption spectroscopy, total reflection Fourier Transform Infrared spectroscopy, and X-ray photoelectron spectroscopy to understand the C-H bonding on the diamond surface.

varies from generally good to excellent. There appears to be adequate resources available to support the directed energy core competency.

Energy Conversion Core Competency

The energy conversion core competency research is at par with top universities and research institutions. Both intramural and extramural scientists and engineers are highly qualified individuals with noted accomplishments in their field of expertise. There are no risks of the energy conversion core competency not reaching its goals and there are no detrimental gaps in the portfolio.

An identified opportunity is:

• While the research shows strong experimental capability, the modeling and computational aspects of some projects are at times fairly basic. Research could be strengthened through increased efforts in modeling and computation to help guide and interpret experimental work, which may require seeking out additional expertise in modeling and computation as such expertise aligns to the needs of the energy conversion projects.

Power Integration and Architecture Core Competency

In general, the research observed in the power integration and architecture core competency is competitive with state-of-the-art work being done elsewhere. This common thread of underlying competence, knowledge of state of the art, appropriate methods of investigation, and building on discoveries was well displayed at ARL. The research within the power integration and architecture core competency exhibits a solid technical quality and understanding. ARL researchers are competitive with their peers in academia, industry, and government in their realization and understanding of the technical issues and challenges at a fundamental science level.

Their contacts and projects with academia, industry, and government appear to have the right mix of partners and issues to keep them and their investigations current. There are no places where the research is at major risk of not meeting its objectives. The facilities supporting the core competency were also found to be excellent.

__________________

^a Core competency descriptions in this passage come from U.S. Army Combat Capabilities Development Command (DEVCOM) Army Research Laboratory (ARL), 2022, “Foundational Research Competencies and Core Competencies,” March.

• The choice of soft actuators is appropriate for some of the morphing projects in the portfolio. However, the field of soft actuation is extensive and rapidly evolving and so the researchers need to continually evaluate their choice of actuation. Advances in direct current motor-drive tendons, fluidic rubber, and fin rays are some areas to consider as projects mature.
• Both analysis and design are noted in projects for aeroelasticity and small air and hypersonic vehicles, and so some advances in multi-disciplinary optimization are relevant. Open-source software that couple optimization and analysis codes in a seamless manner would allow researchers to spend less time on coding and more time on studying the physics of the problem.
• Novel methods for multi-physics of granular materials may be highly beneficial to the research projects by introducing advances in discrete-element modeling coupled with computational fluid dynamics (CFD) and multi-body dynamics.
• The Army Research Laboratory (ARL) could consider doing high-fidelity multi-disciplinary computational modeling to couple with its ongoing experimental efforts for platform design.
• The inclusion of the surrounding environment, such as air quality, sand type, and rock shapes, is a small aspect of the portfolio but one that is of growing importance. In this context, novel methods for multi-physics of granular materials may be highly beneficial to the research projects by introducing advances in discrete-element modeling coupled with CFD and multi-body dynamics.
• The laboratories supporting the competency contain significant equipment that the researchers are effectively using. Researchers with projects on morphing may consider incorporating facilities, such as vibration testing and in-flight deformation analysis, as the research progresses from numerical to experimental. For example, additive manufacturing might be leveraged to advance studies on embodied intelligence with machines like the Carbon MI 3D printer and volumetric additive manufacturing, which will allow printing of structures with which multi-functional materials will interact.

Vehicle Propulsion Sciences Core Competency

The ARL portfolio of research was demonstrative of the very high quality and, for many projects, the leading-edge research conducted by ARL both within its intramural activities and through extramural collaborations. The research was at par with leading institutions nationally and internationally. There do not appear to be risks to the overall portfolio. Without exception, the intramural team at ARL was well qualified for the work they were undertaking and, as was seen across the different competencies reviewed, all of the presenters were smart, well educated, and highly motivated. ARL’s extramural work with leading universities and national laboratories ensure that ARL is working at the theoretical state of the art. The laboratories supporting the core competency contain significant equipment like the sand burner rig and heating microscope that the researchers are effectively using.

Identified opportunities include the following:

• ARL could consider adding engineers and technicians focused on assessing failures and making repairs. Doing so would allow for collection of data on key failure modes and direction of priority research in the future.
• While the scientists at ARL, for the most part, had an understanding of research conducted elsewhere, there were areas identified where a better understanding of recent academic research concerning sand-coating interaction studies and greater connections to industry concerning combustion and ignition could help advance the scientific efforts within this core competency and avoid duplicative efforts. Chapter 5, “Mechanical Sciences,” offers several references on sand-coating interaction studies as well as several suggestions on where researchers may connect to industry.

• In a few projects more experimental and modeling efforts, including those that provide a quantitative analysis, could be incorporated. In other projects, a focus on metrics or a better utilization of data could be added. See Chapter 5, “Mechanical Sciences,” for suggestions on individual projects.
• There are opportunities identified in the Mechanical Sciences Chapter for ARL to advance the development of thermal barrier coatings and environmental barrier coatings, review the chapter for a broader conversation.^b
• An increased focus at a more fundamental level of understanding of calcium-magnesium-aluminosilicate (CMAS) interactions with coatings would benefit the overall program. For example, because the amount, composition, and phase (i.e., crystalline or amorphous) of CMAS applied to thermal environmental barrier coating compositions in furnace testing directly influences the reaction products that form, the CMAS loadings used in furnace tests at ARL need to be carefully considered. ARL researchers may utilize their knowledge from field-tested parts to replicate loadings from field conditions.
• Characterizing CMAS wetting on coating surfaces via a heating microscope is a promising area of research in which ARL is poised to make a positive impact.
• The ARL research on examining solutions to the operation of ground-based diesel engines using fuels with a range of cetane numbers would be a natural extension of the existing studies on flight vehicles. Additionally, there could be more consideration of fuels with a range of octane numbers for vehicles with spark ignition engines.

__________________

^a Core competency descriptions in this passage come from U.S. Army Combat Capabilities Development Command (DEVCOM) Army Research Laboratory (ARL), 2022, “Foundational Research Competencies and Core Competencies,” March.

^b This discussion is relevant to the following presentations shown during the review: “Investigation of Particle Entrainment Glassification of Thermal Barrier Coatings in Gas Turbine Engine Hot Section Components;” “Forensic Investigation of TEBCs on Fielded Engine Turbine Blades;” “Influence of Chemistry and Surface Roughness of Various Thermal Barrier Coating on the Wettability of Molten Sand;” and “Dynamic Wetting of Non-Newtonian Droplets at High Temperatures.”

• It may be beneficial to create a synergistic platform of tools for every project to utilize for their simulation and modeling needs.
• Moving forward it may be important to consider low-power specialized computing resources as larger data sets need to be processed both in real time and as preprocessed large models.
• Achieving data fusion is important and cuts across all research. A first step to accomplish this could include doing a survey of what data sets are most commonly collected in ARL experiments, such as video, images, audio, satellite, underwater, eye tracking data and what kinds of data (e.g., thermal, hyperspectral, or other kinds of sensor data). After understanding the most commonly used forms of data, the investigation could establish which ARL teams are experts in manipulating that kind of data and develop a strategy for more data fusion.^f,g
• While the intramural researchers have a good understanding of the research done outside of ARL, they could gain greater knowledge of the cutting-edge work being done across and within the ARL organization. Several projects internal to ARL could benefit from greater cross-pollination between the intramural researchers. The chapter “Military Information Sciences" (Chapter 6) identifies several opportunities for these potential cross-synergies.
• An important research question that evolved from reviewing these works is “how to understand the limitations of synthetic data?” Such a question could be elevated across the competency as it has ties to non-line-of-sight imaging and counter autonomy.^h,i,j
• Many works presented could benefit from having access to other disparate types of information. What could be helpful is for the researcher to have a vision of what other types of information could be impactful to the specific research being conducted. For example, is there work from weather modeling/data collection that could be useful for research on energy strategies? This kind of “look ahead” focus on data requirements and utility has strong links to data fusion and multi-modal data investigations.
• As more data are collected from a human across a variety of scenarios using virtual reality in simulators, that data can be used to help created educational trainings for personnel in real-world operational situations or to help provide better training materials that are conducive to human information retention and understanding.^k,l
• Many of the project teams could benefit from having a cognitive science expert. The addition of human–computer and human–robot interaction researchers that investigate how best to use humans-in-the-loop for data augmentation, learning, model improvement, and new architecture development could also be helpful.
• An opportunity to connect to the research community beyond publications include recognitions from external organizations such Institute of Electrical and Electronics Engineers (IEEE) awards^m and the Association for the Advancement of Artificial Intelligence awards.ⁿ For AI and robotics, such recognition could come from the IEEE Systems Council awards,^o IEEE Systems, Man, and Cybernetics Society awards,^p or the IEEE Robotics and Automation Society.^q ARL could also consider nominating its researchers to become senior members and fellows of relevant scientific societies, as appropriate.

Reviewing other people’s scholarly works as journal reviewers, and participation in initiatives, such as serving as mentors in student contests would connect researchers to upcoming ideas and state-of-the-art work coming out before works are even published.

Think Faster Research Thrust

Overall, the think faster research thrust is doing relatively high quality work—in some cases, especially the extramural work, it includes leading advances at par with top funding agencies and research institutions; in others, especially with the intramural work, it showed a narrower focus on applications that while not always groundbreaking, clearly relate to unique needs defined by ARL. The researchers are largely aware of research trends in the broader scientific community, although some opportunities to bolster this awareness are listed below. Additionally, ARL was funding top people and universities through its extramural collaborations. The researchers are using sound research methods and methodologies and there are no risks to the overall portfolio.

Identified opportunities include the following:

• AI has made huge advances in the past few years, and the intramural work runs the risk of being out of date soon by sticking to methods that are not being explored at the cutting edge. Some of the performers are indeed cutting-edge researchers, but in a couple of cases for example, the panel knew of newer work, some of it actually performed by the ARL researchers themselves, that were not reported during the review as part of the Army program. This is evidence that ARL researchers have a tremendous resource in each other, and management can facilitate more connections between them for greater cross-pollination of their talents.
• There is an opportunity for some of the world-class expertise of those in listed in the extramural projects supported by Multidisciplinary University Research Initiative (MURIs) to contribute to the specific objectives of many of the intramural projects presented, particularly in the areas of natural language processing (NLP), human–robot interaction, and decision support systems. Leveraging the extramural expertise through deeper connections with the intramural researchers is encouraged.
• ARL collects massive amounts of data for human use. There is an opportunity to leverage this data for automated decision-making with additional research on processing. Thus, one critical direction that could be expanded in the ARL portfolio is multi-modal human–computer interactions (HCI).^r Multi-modal HCI attempts to represent, capture, and communicate the same concept and idea using multiple interaction media or mechanisms. While NLP is a powerful component of this, multi-modal interactions will include other mechanisms such as gesture recognition, voice feature (e.g., pitch, tone) recognition, facial expression recognition, etc. It is well known that NLPs have fundamental difficulties of ambiguity, inadequate vocabulary, etc., which can be ameliorated by leveraging other contextual, possibly non-verbal communication.
• In the fields of artificial intelligence and NLP research, certain terms (e.g., digital twins and formal models) are often used without specific definitions. This phenomenon, endemic to the field at large, was reflected at ARL in how scientists used terms between projects. More interaction between the different researchers working on similar problems could help align the vocabulary that ARL researchers are using.
• There was some research, especially in the area of language, where “simplified” modeling and simulation was used instead of experimentation with actual robots or with the simulators used by the robotics researchers. The reason for this is that there is a well-known gap between the actual physical competencies of robots, both in sensing and action, and the linguistic capabilities that would be desired by those controlling them. It is suggested that military information systems competency leadership work more closely with ARL experts in the Robotics Research Collaboration Campus to explore how this gap might be filled.
• Some of the scientific questions that might be most useful to the military information systems competency are being explored in other parts of the AI world (e.g., distributed learning work is aiming to scale to significant levels in the presence of limited bandwidth,

such as would be needed in autonomous vehicles).^s The competency could also consider casting a wider net on the modern AI front and could also make sure that opportunities coming up in these other domains are more scalable. Related AI problems may be unintentionally missed by the Army because of overly narrow definitions of adversarial reasoning and causal modeling. Two good examples of this are hybrid utility models^t and the use of graph networks (and graph neural network learning) in representing adversarial problems.^u,v,w References are supplied in the footnotes below on these topics that may be helpful to ARL.

• The work in abstract meaning representation was being used for directing robots, but not directly in the bidirectional NLP work. The latter provides an essential future capability for robots deployed in novel environments where there may not be a lot of training data. Closer cooperation between internal teams might be useful.
• The poster “Neuro-Symbolic AI/ML for Complex Event Processing” shows intramural research focused on integrating symbolic and neural AI approaches. Those approaches in NLP are an emerging area of importance in AI research, and ARL could explore how the laboratory’s expertise in reasoning under uncertainty, adversarial reasoning, and resource-limited reasoning might be integrated with the work in NLP, which is being pursued separately.
• The competency seems well resourced in terms of computational power and access to simulation to support the think faster research thrust. It was less clear, however, whether access to the actual robots was available and whether the interactions of this topic with the overall robotics work was realistic. For example, some of the NLP demonstrations included interactions with simulated robots that were not very realistic. This was useful for the particular linguistic work they were demonstrating, but would not lead to realistic human-robot interaction in the short term, unless more work was done interacting with the detailed simulators or actual robots that the Army has available in its research facilities.

Act Smarter Research Thrust Area

The overall presented work in the act smarter research thrust area was clear, compelling, and at par with leading funding agencies nationally and internationally. While several researchers exhibited that they had good knowledge of other research, this was not the case across the board and some of the presentations would be improved with specific content positioning the project’s research in the state-of-the-art literature. The overarching research questions identified within the project presentations are very good, and represent tough challenges. Most presenters had published their work in top venues. It is commendable that four of the five planned demos worked and that the teams had backup plans in place when glitches occurred. ARL is making good choices in which extramural projects to emphasize and which to de-emphasize moving forward. The ARL researchers supporting the act smarter research thrust appear to be well qualified to perform the research they are undertaking.

Demos were managed capably, and, as mentioned, success in four out of five research-class experiments is commendable. Grace’s Quarter offers appropriate test environments for ARL robots. ARL demonstrations successfully utilized UGV, UAS, and quadruped robots in simulated missions. These robots were well chosen and appropriate for each assigned task. Control stations were well equipped, and large monitors were available for onsite briefings.

Identified opportunities include the following:

• There appear to be several opportunities in cross-project collaboration. In general, efforts to better integrate theoretical, computational, and hardware-based projects need to be identified and pursued when possible. For example, maps from the “Geospatial Data Integration Server” project need to be utilized in Terrain-Aware Autonomous Ground Navigation to maximize situational awareness beyond sensor line-of-sight, minimize the need for real-time simultaneous localization and mapping (SLAM), and facilitate feedback of every deployed robot SLAM data set into the Geospatial Data Server. As another example, collaboration between adaptive planner parameter learning and real-world

artificial intelligence of maneuver and mobility in off-road autonomy trials will enable capture of human expert data from Graces’ Quarters experiments, which in turn will offer improved expert-informed planning autonomy for fielded robots.

• There are three traditional disciplines within the act smarter project group: human-machine systems (e.g., human expert–guided machine learning [ML]), multi-agent coordination (e.g., swarms), and experimental robotic platform development and field-testing. Creativity will be enhanced by bringing ARL personnel with diverse mindsets together—for example, human-guided planning and cooperative control mathematics experts. Similarly, there will need to be more coordination of the talents between the researchers in the three research thrust areas to inspire more innovation. For example, the act smarter presentation on planning only focused on motion planning without consideration of the greater mission and its modeling and planning. This is because there is one “discipline silo” for mission task planning (in the think faster research thrust area) and another “discipline silo” for acting (the act smarter thrust area). There are opportunities for mission and tactical (motion) planner integration that will be better realized through these groups working together. Similarly, there is unrealized collaboration potential with the perception team working on sensor data processing to “map” an environment (the “M” in SLAM).
• The extramural and intramural managers at ARL could regularly interact to build and maintain a synergistic research portfolio. Substantial efficiency gains may be possible by minimizing overlap and maximizing awareness of both portfolios. Products such as data and software for non-Controlled Unclassified Information efforts can then be best shared and cooperatively grown.
• ARL may consider upscaling the multi-agent team sizes. Since the act smarter research thrust emphasizes multi-agent teams, it is critical to deploy teams capable of exposing any flaws in theoretical and simulation studies related to team size. Such experiments offer opportunities to identify and document planned and unanticipated interactions in real world testing. Adding resources to acquire, maintain, and deploy larger quantities of robots for multi-agent experiments is suggested. The team could also address challenges in scaling up user interfaces (i.e., the number of people and the number of control stations) to support multi-agent testing.
• It was noted that most robots were off-the-shelf, suggesting a need for more manufacturing capability to support customization of hardware. While the robot chassis used at ARL are mostly off-the-shelf, it is apparent that significant effort is required to mount sensors and electronics. Particularly for those robots operating in rugged outdoor environments that require cooling and shock mounting, there are few commercial solutions. Additionally, some of the ARL researchers asked for more convenient access to machine shops; this is an important asset for rapid prototyping of experimental vehicles. Such machine shops could be used to hang sensors, build pan and tile mounts, add additional shock enclosures for electronics, as well as for additive manufacturing and electronics fabrication.
• Collaborations between ARL and experts in academia could be enhanced by placing ARL personnel in academic institutions. A complementary mutually beneficial approach could involve academic partners working at ARL facilities.
• To ensure user acceptance and build truly trustworthy decision-support systems and robotic systems requires additional functionality, including explanation capabilities. Autonomous “agents” (i.e., robots and software packages) that make decisions need to be able to explain their decisions to other agents (human or robotic). Explanations may be graphical, numerical, NLP-based, or a combination.
• Explanation capabilities are currently understudied at ARL. Centralizing basic research on them could be supportive to individual subprojects. For example, a swarm should be able to explain how it is being deceptive with its motions; a planner should be able to explain how a new decision is guided by what an expert previously recommended; or a deployed vehicle that has been out of contact should be able to explain how and why it chose to go

silent and how and why it re-established contact.

• Achieving robustness is an important scientific topic that needs to be emphasized even more across the research portfolio of ARL. The Army frequently exposes robotic platforms to high-risk and adversarial environments, and each robot must be able to succeed and even thrive despite failures and fog-of-war conditions. The principles of robust feedback control can help individual robots continue functioning despite systems failures.
  • ML can improve robustness to changes in models or environmental conditions.
  • Multi-agent systems can be made robust with task re-allocation and recovery capabilities should robotic team members be damaged or lost.
  • Diversity can also improve robustness. There is a promising pathway toward this capability with coordinated missions involving combinations of air, ground, and legged robots.
  • There is more to explore in heterogeneous large-scale multi-agent and human–robot team planning and coordination research.

__________________

^a U.S. Army Combat Capabilities Development Command (DEVCOM) Army Research Laboratory (ARL), 2022, “Foundational Research Competencies and Core Competencies,” March.

^b DEVCOM ARL, 2023, “Military Information Sciences Competency Story: Operationalizing Science for an Army That Looks Farther, Thinks Faster and Acts Smarter,” August 25.

^c D. Wold, 1996, “JOVE Final Report,” University of Arkansas at Little Rock Department of Physics and Astronomy, https://ntrs.nasa.gov/api/citations/19970041452.

^d State-of-the-art methods refers to the use most innovative and best performing models, techniques, algorithms, and technologies in ML.

^e R. Chandra et al., 2020, “Forecasting Trajectory and Behavior of Road-Agents Using Spectral Clustering in Graph-LSTMs,” IEEE Robotics and Automation Letters 5(3): 4882–4890, https://doi.org/10.1109/LRA.2020.3004794.

^f M. Schmitt and X. Zhu, 2016, “Data Fusion and Remote Sensing: An Ever-Growing Relationship,” IEEE Geoscience and Remote Sensing Magazine 4, December, https://doi.org/10.1109/MGRS.2016.256102.

^g J. Gawlikowski, S. Saha, J. Niebling, and X.X. Zhu, 2022, “Robust Distribution-Shift Aware Sar-Optical Data Fusion for Multi-Label Scene Classification,” 2022 IEEE International Geoscience and Remote Sensing Symposium, https://doi.org/10.1109/IGARSS46834.2022.9884880.

^h Y. Tian, O.J. Hénaff, and A.V.D. Oord, 2021, “Divide and Contrast: Self-Supervised Learning from Uncurated Data,” 2021 IEEE/CVF International Conference on Computer Vision (ICCV), https://doi.org/10.1109/ICCV48922.2021.00991.

ⁱ H. Kuang et al., 2021, “Video Contrastive Learning with Global Context,” IEEE/CVF International Conference on Computer Vision Workshops (ICCVW), https://doi.org/10.1109/ICCVW54120.2021.00358.

^j K. Kotar, G. Ilharco, L. Schmidt, K. Ehsani, and R. Mottaghi, 2021, “Contrasting Contrastive Self-Supervised Representation Learning Pipelines,” 2021 ICCV, https://doi.org/10.1109/ICCV48922.2021.00980.

^k A. Lekova, P. Tsvetkova, and A. Andreeva, 2023, “System Software Architecture for Enhancing Human-Robot Interaction by Conversational AI,” 2023 International Conference on Information Technologies (InfoTech), https://doi.org/10.1109/InfoTech58664.2023.10266870.

^l D. Pham et al., 2022, “A Case Study of Human-AI Interactions Using Transparent AI-Driven Autonomous Systems for Improved Human-AI Trust Factors,” IEEE 3rd International Conference on Human-Machine Systems (ICHMS), https://doi.org/10.1109/ICHMS56717.2022.9980662.

^m IEEE, https://corporate-awards.ieee.org, accessed February 4, 2023.

ⁿ AAAI, https://aaai.org/about-aaai/aaai-awards, accessed February 4, 2023.

^o IEEE System Council, https://ieeesystemscouncil.org/awards, accessed February 4, 2023.

^p IEEE Systems, Man, and Cybernetics Society, https://www.ieeesmc.org/about-smc/awards, accessed February 4, 2023.

^q IEEE Robotics & Automation Society, https://www.ieee-ras.org/awards-recognition, accessed February 4, 2023.

^r A. Holzinger, B. Malle, A. Saranti, and B. Pfeifer, 2021, “Towards Multi-Modal Causability with Graph Neural Networks Enabling Information Fusion for Explainable AI,” Information Fusion 71, https://doi.org/10.1016/j.inffus.2021.01.008.

^s D. Otto, G. Scharnberg, M. Kerres, and O. Zawacki-Richter, eds., 2023, Distributed Learning Ecosystems: Concepts, Resources, and Repositories, Springer Nature.

^t Z. Zhao et al., 2020, “Learning Mixtures of Random Utility Models with Features from Incomplete Preferences,” https://doi.org/10.48550/arXiv.2006.03869.

^u V. Hassija et al., 2020, “DAGIoV: A Framework for Vehicle to Vehicle Communication Using Directed Acyclic

Graph and Game Theory,” IEEE Transactions on Vehicular Technology, January, https://doi.org/10.1109/TVT.2020.2968494.

^v Z. Bu et al., 2019, “Link Prediction in Temporal Networks: Integrating Survival Analysis and Game Theory,” Information Sciences 498, https://doi.org/10.1016/j.ins.2019.05.050.

^w A. Duval and F.D. Malliaros, 2021, “GraphSVX: Shapley Value Explanations for Graph Neural Networks,” in Machine Learning and Knowledge Discovery in Databases. Research Track (N. Oliver, F. Pérez-Cruz, S. Kramer, J. Read, and J.A. Lozano, eds.), ECML PKDD 2021. Lecture Notes in Computer Science, 12976. Springer, Cham., https://doi.org/10.1007/978-3-030-86520-7_19.