Neural Radiance Fields (NeRFs) have been remarkably successful at synthesizing novel views of 3D scenes by optimizing a volumetric scene function. This scene function models how optical rays bring color information from a 3D object to the camera pixels. Radio frequency (RF) or audio signals can also be viewed as a vehicle for delivering information about the environment to a sensor. However, unlike camera pixels, an RF/audio sensor receives a mixture of signals that contain many environmental reflections (also called “multipath”). Is it still possible to infer the environment using such multipath signals? We show that with redesign, NeRFs can be taught to learn from multipath signals, and thereby “see” the environment. As a grounding application, we aim to infer the indoor floorplan of a home from sparse WiFi measurements made at multiple locations inside the home. Although a difficult inverse problem, our implicitly learnt floorplans look promising, and enables forward applications, such as indoor signal prediction and basic ray tracing.

The SEND/RETURN (S/R) project is created to explore the efficacy of content-based music recommendations alongside a uniquely generated Unreal Engine 5 (UE5) virtual environment based on audio features. S/R employs both a k-means clustering algorithm using audio features and a fast pattern matching (FPM) algorithm using 30-second audio signals to find similar-sounding songs to recommend to users. The feature values of the recommended song are then communicated via HTTP to the UE5 virtual environment, which changes a number of effects in real-time. All of this is being replicated from a listen-server to other clients to create a multiplayer audio session. S/R successfully creates a lightweight online environment that replicates song information to all clients and suggests new songs that alter the world around you. In this work, we extend S/R by training a convolutional neural network using Mel-spectrograms of 30-second audio samples to predict the mood of a song. This model can then orchestrate the post-processing effect in the UE5 virtual environment. The developed convolutional model had a validation accuracy of 67.5% in predicting 4 moods ('calm', 'energetic', 'happy', 'sad').

Despite an increasing number of successful interventions designed to broaden participation in computing research, there is still significant attrition among historically marginalized groups in the computing research pipeline. This experience report describes a first-of-its-kind Undergraduate Consortium (UC; https://aaai-uc.github.io/about) that addresses this challenge by empowering students with a culmination of their undergraduate research in a conference setting. The UC, conducted at the AAAI Conference on Artificial Intelligence (AAAI), aims to broaden participation in the AI research community by recruiting students, particularly those from historically marginalized groups, supporting them with mentorship, advising, and networking as an accelerator toward graduate school, AI research, and their scientific identity. This paper presents our program design, inspired by a rich set of evidence-based practices, and a preliminary evaluation of the first years that points to the UC achieving many of its desired outcomes. We conclude by discussing insights to improve our program and expand to other computing communities.

Audio adversarial attacks on deep learning models are of great interest given the commercial success and proliferation of these technologies. These types of attacks have been successfully demonstrated, however, artifacts introduced in the adversarial audio are easily detectable by a human observer. In this work, an expansion of the fast audio adversarial perturbation framework is proposed that can produce an adversarial attack that is imperceptible to a human observer in near-real time using black-box attacks. This is achieved by proposing a perceptually motivated penalty function. We propose a perceptual fast audio adversarial perturbation generator (PFAPG) that employs a loudness constrained loss function, in lieu of a conventional L-2 norm, between the adversarial example and original audio signal. We compare the performance of PFAPG against the conventional constraint based on the MSE on three audio recognition datasets: speaker recognition, speech command, and the Ryerson audiovisual database of emotional speech and song. Our results indicate that, on average, PFAPG equipped with the loudness-constrained loss function yields a 11% higher success rate, while reducing the undesirable distortion artifacts in adversarial audio by 10% dB compared to the prevalent MSE constraints.

In this work, we introduce and evaluate federated fine-tuning (FFT) toward developing decentralized systems for IoT applications using edge computing. We demonstrate a deployed, off-grid, FFT network composed of embedded hardware and assess the system and its performance. The federated averaging algorithm has become the popular approach in decentralized systems with multiple nodes due its low computational costs and simplicity. However, it is commonly implemented using a workstation or a cloud as its server node, typically demonstrated with unrealistic (small) neural network models and may have high communication cost for embedded applications. To address these challenges, we present two main contributions by: (1) Improving the federated averaging algorithm’s weight initialization step and by limiting percentage of weights being averaged to enhance system security and model performance, and (2) Demonstrating the proposed system’s effectiveness via deployment of realistic model using an edge device as the server node (first time for a federated system). For our evaluation, we use a centrally pre-trained MobileNetV2 model on the CelebA dataset. We record the transmitted model parameters across the network with the modified federated averaging algorithm and FFT, and capture metrics related to memory, power consumption, CPU load, and communication on the device. Overall, results demonstrate that FFT can improve federated system performance and model accuracy while providing stronger privacy, protection of intellectual property, and security against adversarial attacks on federated learning.

This paper introduces a novel hybrid evolutionary algorithm to solve the 2019 IEEE CEC Competition on 100-Digit Challenge on Single Objective Numerical Optimization. The proposed algorithm, named GADE, employs a genetic algorithm (GA) to explore the search space. If a complete solution is not found with GA, differential evolution (DE) exploits the search space using the latest GA solution candidates. We present two techniques to improve GA's exploration capabilities: fitness-based opposition and tidal mutation. Simulations on the ten challenge problems indicate that fitness-based opposition allows more GA simulations to find the correct digits. Also while GA and DE can fully solve four of the problems on their own, the hybrid algorithm allows for higher scores in at least three of the remaining problems. Furthermore, we provide analysis on the contribution of each EA to the score based on their cost function evaluations.

Artificial intelligence (AI) and machine learning are rapidly developing fields, garnering increased attention from the public, as well as academia and industry. Recent advances allow us to use our smart devices at home to control TVs, speakers, and lights. A smart speaker that is plugged into the wall can handle powerful processing that can service complex user requests. However, as we rely more and more on AI conveniences, we expect them to be at our disposal when we are moving around. In this poster, we present preliminary work on the development of a new upper-level undergraduate course open to all computer science, electrical engineering, and applied mathematics students at Wentworth Institute of Technology (WIT). The new course, titled Embedded AI (EAI), introduces topics from embedded systems, artificial intelligence, digital signal processing, linear algebra, and probability and statistics. The proposed course includes lectures introducing the theory behind each topic and laboratory assignments that provide the students with the hands-on experience. At the end of the semester, students develop an embedded AI project of their own. An interdisciplinary academic team of computer scientists and electrical and computer engineers has been working on a curriculum and researching different platforms, datasets and sensors that will ultimately enable students at WIT to develop an embedded device that can support AI tasks as well as give the students the practical skills that employers seek. Handouts will be prepared with sample curriculum and breakdown of topics.

Evolutionary algorithms (EAs) are widely employed for solving optimization problems with rugged fitness landscapes. Opposition-based learning (OBL) is a recent tool developed to improve the convergence rate of EAs. In this paper, we derive the probabilities that distances between OBL points and the optimization problem solution are less than the distance between a given EA individual and the optimal solution. We find that the quasi-reflected opposition point yields the highest probability and is the most likely candidate to be closer to the optimal solution. We then employ CEC 2013 competition benchmark problems and select a set of trajectory optimization problems from the European Space Agency to study the performance of three OBL algorithms in conjunction with three different EAs. The CEC 2013 test suit simulations indicate that quasi-reflection accelerates the performance of the EA, especially for more difficult composition functions. The space trajectory experiments reveal that differential evolution with opposition generally returns the best objective function value for the chosen minimization problems.

In this paper, we propose a framework for employing opposition-based learning to assist evolutionary algorithms in solving discrete and combinatorial optimization problems. To our knowledge, this is the first attempt to apply opposition to combinatorics. We introduce two different methods of opposition to solve two different type of combinatorial optimization problems. The first technique, open-path opposition, is suited for combinatorial problems where the final node in the graph does not have be connected to the first node, such as the graph coloring problem. The latter technique, circular opposition, can be employed for problems where the endpoints of a graph are linked, such as the well-known traveling salesman problem (TSP). Both discrete opposition methods have been hybridized with biogeography-based optimization (BBO). Simulations on TSP benchmarks illustrate that incorporating opposition into BBO improves its performance.

Biogeography-based optimization (BBO) is a population-based evolutionary algorithm that is based on the mathematics of biogeography. Biogeography is the science and study of the geographical distribution of biological organisms. In BBO, problem solutions are analogous to islands, and the sharing of features between solutions is analogous to the migration of species. This paper derives Markov models for BBO with selection, migration, and mutation operators. Our models give the theoretically exact limiting probabilities for each possible population distribution for a given problem. We provide simulation results to confirm the Markov models.

Biogeography-based optimization (BBO) is a recently developed heuristic algorithm which has shown impressive performance on many well known benchmarks. In order to improve BBO, this paper incorporates distinctive features from other successful heuristic algorithms into BBO. In this paper, features from evolutionary strategy (ES) are used for BBO modification. Also, a new immigration refusal approach is added to BBO. After the modification of BBO, F-tests and T-tests are used to demonstrate the differences between different implementations of BBOs.

We propose a novel variation to biogeography-based optimization (BBO), which is an evolutionary algorithm (EA) developed for global optimization. The new algorithm employs opposition-based learning (OBL) alongside BBO's migration rates to create oppositional BBO (OBBO). Additionally, a new opposition method named quasi-reflection is introduced. Quasi-reflection is based on opposite numbers theory and we mathematically prove that it has the highest expected probability of being closer to the problem solution among all OBL methods. The oppositional algorithm is further revised by the addition of dynamic domain scaling and weighted reflection. Simulations have been performed to validate the performance of quasi-opposition as well as a mathematical analysis for a single-dimensional problem. Empirical results demonstrate that with the assistance of quasi-reflection, OBBO significantly outperforms BBO in terms of success rate and the number of fitness function evaluations required to find an optimal solution.

Biogeography-based optimization (BBO) is an evolutionary algorithm that is based on the science of biogeography. Biogeography is the study of the geographical distribution of organisms. In BBO, problem solutions are represented as islands, and the sharing of features between solutions is represented as migration between islands. This paper develops a Markov analysis of BBO, including the option of elitism. Our analysis gives the probability of BBO convergence to each possible population distribution for a given problem. We compare our BBO Markov analysis with a similar genetic algorithm (GA) Markov analysis. Analytical comparisons on three simple problems show that with high mutation rates the performance of GAs and BBO is similar, but with low mutation rates BBO outperforms GAs. Our analysis also shows that elitism is not necessary for all problems, but for some problems it can significantly improve performance.