Accurate forecasts are essential for integrating wind energy into the power grid. With wind energy's growing role in the renewable mix, precise short-term generation forecasts are increasingly vital. Turbine-level forecasts are critical for optimal wind farm operation, control, and planning. However, the wind's unpredictability and the complex interactions between turbines make forecasting challenging. Existing methods either depend on resource-intensive physical models or overlook turbine interactions, resulting in isolated, overly simplistic models.

This master's thesis explores the use of Graph Neural Networks (GNNs) in conjunction with Recurrent Neural Networks (RNNs) for wind power generation forecasting in wind farms. The result is an integrated approach that can learn the interactions between turbines through GNNs to handle the temporal dynamics of the data through RNNs.

The research begins with an in-depth overview of wind turbines and their arrangement in wind farms, highlighting the importance and challenges of Wind Power Forecasting (WPF). It establishes that the spatial configuration of turbines complicates power generation estimation, especially over different time horizons. Graphs are introduced as an effective way to represent the interconnections between turbines. Various methods for constructing these graphs are reviewed, focusing on predefined heuristics and their limitations in adapting to changing wind farm characteristics.

The core of this thesis is the AG-LSTM Network, a model that integrates RNNs and GNNs for short-term WPF for turbines within a wind farm. It utilizes historical power generation data, a variety of future and past covariates, and static turbine-specific features. The model is based on an encoder-decoder architecture with an Adaptive Graph Long-Short Term Memory Cell (AG-LSTM), which generates a dynamic adjacency matrix representing the farm's state at each moment. This matrix combines the input features and hidden states across turbines, resulting in an embedding that is transformed into the power forecast.

Through comprehensive experiments and analyses, the AG-LSTM Network proves to be more accurate than state-of-the-art methods on two real-life datasets when future covariates are unavailable. Further experiments demonstrate the quality of the forecast and break down the contribution of the different choices regarding its architecture. The results underscore the potential of hybrid neural network architectures in enhancing WPF, providing a valuable tool for the renewable energy sector.

In conclusion, this research contributes to the renewable energy field by proposing a novel hybrid neural network model that effectively addresses the complexities of short-term WPF. Future work could explore further exploitation of future covariates and the robustness of the model against incomplete input data.

Modular neural networks have received an upsurge of attention lately owing to their unique modular design and potential capacity to decompose complex dynamics and learn interactions among causal variables. Inspired by this potential, we employ the recently introduced Recurrent Independent Mechanisms (RIMs) in the downstream video prediction task. RIMs consist of several modular recurrent units and modular hidden states which are called RIM cells. Those modules are connected by two attention mechanisms. Through experiments, we show that RIMs perform better or comparably with related baselines.

From modular recurrent units to modular image representations, we push the modularity further to explore how much the performance can benefit from it. We extend RIMs architecture on both the encoder and decoder sides to allow for object-centric (OC) feature representation learning in video prediction, resulting in an end-to-end architecture we refer to as OC-RIMs. Our qualitative evaluations demonstrate that every RIM cell in OC-RIMs now attends to a certain object within the input scene at any specific moment. As a result, OC-RIMs offer considerable quantitative performance improvement in video prediction over comparable baselines across two datasets.

We perform extensive ablation studies to validate the design choices of every module of RIMs. We empirically show that most modules work as expected. However, the sparse activation greatly detriments the prediction performance, which is against the claims in the paper where RIMs were proposed. On the other hand, RIM cells are expected to work near-independently. But experiments show that the use of communication mechanism leads to heavy co-adaptation between cells, i.e., RIM cells fail to make any reasonable predictions independently. Those issues have raised our concerns about the design of RIMs. Finally, we point out some future work directions to address these deficiencies.

Typically, people search images by text: users enter keywords and a search engine returns relevant results. However, this pattern has limitations. An obvious drawback is that when searching in one language, users may miss results labelled in other languages. Moreover, sometimes people know little about the object in the image and thus would not know what keywords could be used to search for more information. Driven by this use case with many applications, content-based image retrieval (CBIR) has recently been put under the spotlight, which aims to retrieve similar images in the database solely by the content of the query image without relying on textual information.
To achieve this objective, an essential part is that the search engine should be able to interpret images at a higher level instead of treating them simply as arrays of pixel values. In practice, this is done by extracting distinguishable features. Many effective algorithms have been proposed, from traditional handcrafted features to more recent deep learning methods. Good features may lead to good retrieval performance, but the problem is still not fully solved. To make the engine useful in real-world applications, retrieval efficiency is also an important factor to consider while has not received as much attention as feature extraction.
In this work, we focus on retrieval efficiency and provide a solution for real-time CBIR in million-scale databases. The feature vectors of database images are extracted and stored offline. During the online procedure, such feature vectors of query images are also extracted and then compared with database vectors, finding the nearest neighbours and returning the corresponding images as results. Since feature extraction only performs once for each query, the main limiting factor of retrieval efficiency in large-scale database is the time of finding nearest neighbours. Exact search has been shown to be far from adequate, and thus approximate nearest neighbour (ANN) search methods have been proposed, which mainly fall into two categories: compression-based and tree/graph-based. However, these two types of approaches are usually not discussed and compared together. Also, the possibility of combining them has not been fully studied. Our study (1) applies and compares methods in both categories, (2) reveals the gap between toy examples and real applications, and (3) explores how to get the best of both worlds. Moreover, a prototype of our image search engine with GUI is available on https://github.com/YYao-42/ImgSearch.

The underwater acoustic environment is amongst the most challenging mediums for wireless communications. The three distinct challenges of underwater acoustic communication are the low and nonuniform propagation speed, frequency-dependent attenuation and time-varying multipath propagation.
To cope with these challenges, physical layer communication systems allow the selection of communication parameters based on environmental conditions and constraints. This is also known as link adaptation. In this thesis, the frequency-repetition spread-spectrum (FRSS) physical layer is studied. Various channel parameters are used to classify the optimal FRSS format. Furthermore, different machine learning classifiers are implemented to solve the classification problem. It is determined that the output signal-to-noise ratio provides enough information to switch effectively between transmission formats. Among the implemented machine learning classifiers, the decision tree strikes a good balance between performance and computational complexity. It is shown that a small performance gain can be achieved when custom channel parameters are extracted from the estimated impulse response and the equalizer error sequence using a deep neural network.
Optimizing the equalization process is another method to better cope with difficult environmental conditions. Various adaptive filter algorithms are implemented for the decision feedback equalizer used in the FRSS receiver. The optimal algorithm parameters are found by means of algorithm unrolling. It is shown that the standard least mean squares algorithm cannot be outperformed by various other optimization algorithms that use linear or non-linear filters.
The Watermark channel simulator is used to study the performance of the link adaptation and equalization optimization solutions for a wide range of underwater channels.