updated links in the papers

content/publications/pub0.md

@@ -15,3 +15,7 @@ post_meta = ["author","categories", "translations"]
 
 ## Abstract
 A thorough regulation of building energy systems translates in relevant energy savings and in a better comfort for the occupants. Algorithms to predict the thermal state of a building on a certain time horizon with a good confidence are essential for the implementation of effective control systems. This work presents a global Transformer architecture for indoor temperature forecasting in multi-room buildings, aiming at optimizing energy consumption and reducing greenhouse gas emissions associated with HVAC systems. Recent advancements in deep learning have enabled the development of more sophisticated forecasting models compared to traditional feedback control systems. The proposed global Transformer architecture can be trained on the entire dataset encompassing all rooms, eliminating the need for multiple room-specific models, significantly improving predictive performance, and simplifying deployment and maintenance. Notably, this study is the first to apply a Transformer architecture for indoor temperature forecasting in multi-room buildings. The proposed approach provides a novel solution to enhance the accuracy and efficiency of temperature forecasting, serving as a valuable tool to optimize energy consumption and decrease greenhouse gas emissions in the building sector.
+
+[[paper]()]
+[[arXiv](https://arxiv.org/abs/2310.20476)]
+[[github]()]

content/publications/pub1.md

@@ -16,3 +16,7 @@ post_meta = ["categories", "translations"]
 
 ## Abstract
 Time series forecasting is an important problem, with many real world applications. Transformer models have been successfully applied to natural language processing tasks, but have received relatively little attention for time series forecasting. Motivated by the differences between classification tasks and forecasting, we propose PI-Transformer, an adaptation of the Transformer architecture designed for time series forecasting, consisting of three parts: First, we propose a novel initialization method called Persistence Initialization, with the goal of increasing training stability of forecasting models by ensuring that the initial outputs of an untrained model are identical to the outputs of a simple baseline model. Second, we use ReZero normalization instead of Layer Normalization, in order to further tackle issues related to training stability. Third, we use Rotary positional encodings to provide a better inductive bias for forecasting. Multiple ablation studies show that the PI-Transformer is more accurate, learns faster, and scales better than regular Transformer models. Finally, PI-Transformer achieves competitive performance on the challenging M4 dataset, both when compared to the current state of the art, and to recently proposed Transformer models for time series forecasting.
+
+[[paper](https://link.springer.com/article/10.1007/s10489-023-04927-4)]
+[[arXiv](https://arxiv.org/abs/2208.14236)]
+[[github]()]

content/publications/pub10.md

@@ -13,3 +13,7 @@ post_meta = ["categories", "translations"]
 
 ## Abstract
 This letter investigates the potential of the electrocardiogram to perform early meal detection, which is critical for developing a fully-functional automatic artificial pancreas. The study was conducted in a group of healthy subjects with different ages and genders. Two classifiers were trained: one based on neural networks (NNs) and working on features extracted from the signals and one based on convolutional NNs (CNNs) and working directly on raw data. During the test phase, both classifiers correctly detected all the meals, with the CNN outperforming the NN in terms of misdetected meals and detection time (DT). Reliable meal onset detection with short DT has significant practical implications: It reduces the risk of postprandial hyperglycemia and hypoglycemia, and it reduces the mental burden of meal documentation for patients and related stress.
+
+[[paper](https://ieeexplore.ieee.org/document/10225027)]
+[[arXiv]()]
+[[github]()]

content/publications/pub11.md

@@ -13,3 +13,7 @@ post_meta = ["categories", "translations"]
 
 ## Abstract
 We present a novel time series anomaly detection method that achieves excellent detection accuracy while offering a superior level of explainability. Our proposed method, TimeVQVAE-AD, leverages masked generative modeling adapted from the cutting-edge time series generation method known as TimeVQVAE. The prior model is trained on the discrete latent space of a time-frequency domain. Notably, the dimensional semantics of the time-frequency domain are preserved in the latent space, enabling us to compute anomaly scores across different frequency bands, which provides a better insight into the detected anomalies. Additionally, the generative nature of the prior model allows for sampling likely normal states for detected anomalies, enhancing the explainability of the detected anomalies through counterfactuals. Our experimental evaluation on the UCR Time Series Anomaly archive demonstrates that TimeVQVAE-AD significantly surpasses the existing methods in terms of detection accuracy and explainability.
+
+[[paper]()]
+[[arXiv](https://arxiv.org/abs/2311.12550)]
+[[github]()]

content/publications/pub2.md

@@ -14,3 +14,7 @@ thumbnail = "images/maze.jpeg"
 
 ## Abstract
 The field of time series anomaly detection is constantly advancing, with several methods available, making it a challenge to determine the most appropriate method for a specific domain. The evaluation of these methods is facilitated by the use of metrics, which vary widely in their properties. Despite the existence of new evaluation metrics, there is limited agreement on which metrics are best suited for specific scenarios and domain, and the most commonly used metrics have faced criticism in the literature. This paper provides a comprehensive overview of the metrics used for the evaluation of time series anomaly detection methods, and also defines a taxonomy of these based on how they are calculated. By defining a set of properties for evaluation metrics and a set of specific case studies and experiments, twenty metrics are analyzed and discussed in detail, highlighting the unique suitability of each for specific tasks. Through extensive experimentation and analysis, this paper argues that the choice of evaluation metric must be made with care, taking into account the specific requirements of the task at hand.
+
+[[paper](https://link.springer.com/article/10.1007/s10618-023-00988-8)]
+[[arXiv](https://arxiv.org/abs/2303.01272)]
+[[github](https://github.com/sondsorb/tsad_eval)]

content/publications/pub3.md

@@ -15,3 +15,6 @@ thumbnail = "images/towers.png"
 ## Abstract
 Accurately forecasting traffic in telecommunication networks is essential for operators to efficiently allocate resources, provide better services, and save energy. We propose Circle Attention, a novel spatial attention mechanism for telecom traffic forecasting, which directly models the area of effect of neighboring cell towers. Cell towers typically point in three different geographical directions, called sectors. Circle Attention models the relationships between sectors of neighboring cell towers by assigning a circle with learnable parameters to each sector, which are: the azimuth of the sector, the distance from the cell tower to the center of the circle, and the radius of the circle. To model the effects of neighboring time series, we compute attention weights based on the intersection of circles relative to their area. These attention weights serve as multiplicative gating parameters for the neighboring time series, allowing our model to focus on the most important time series when making predictions. The circle parameters are learned automatically through back-propagation, with the only signal available being the errors made in the traffic forecasting of each sector. To validate the effectiveness of our approach, we train a Transformer to forecast the number of attempted calls to sectors in the Copenhagen area, and show that Circle Attention outperforms the baseline methods of including either all or none of the neighboring time series. Furthermore, we perform an ablation study to investigate the importance of the three learnable parameters of the circles, and show that performance deteriorates if any of the parameters are kept fixed. Our method has practical implications for telecommunication operators, as it can provide more accurate and interpretable models for forecasting network traffic, allowing for better resource allocation and improved service provision.
 
+[[paper](https://dl.acm.org/doi/10.1007/978-3-031-43430-3_7)]
+[[arXiv]()]
+[[github]()]

content/publications/pub4.md

@@ -15,3 +15,7 @@ post_meta = ["categories", "translations"]
 
 ## Abstract
 This paper presents a novel sampling scheme for masked non-autoregressive generative modeling. We identify the limitations of TimeVQVAE, MaskGIT, and Token-Critic in their sampling processes, and propose Enhanced Sampling Scheme (ESS) to overcome these limitations. ESS explicitly ensures both sample diversity and fidelity, and consists of three stages: Naive Iterative Decoding, Critical Reverse Sampling, and Critical Resampling. ESS starts by sampling a token set using the naive iterative decoding as proposed in MaskGIT, ensuring sample diversity. Then, the token set undergoes the critical reverse sampling, masking tokens leading to unrealistic samples. After that, critical resampling reconstructs masked tokens until the final sampling step is reached to ensure high fidelity. Critical resampling uses confidence scores obtained from a self-Token-Critic to better measure the realism of sampled tokens, while critical reverse sampling uses the structure of the quantized latent vector space to discover unrealistic sample paths. We demonstrate significant performance gains of ESS in both unconditional sampling and class-conditional sampling using all the 128 datasets in the UCR Time Series archive.
+
+[[paper]()]
+[[arXiv](https://arxiv.org/abs/2309.07945)]
+[[github]()]

content/publications/pub5.md

@@ -13,3 +13,7 @@ post_meta = ["categories", "translations"]
 
 ## Abstract
 This paper presents a novel sampling scheme for masked non-autoregressive generative modeling. We identify the limitations of TimeVQVAE, MaskGIT, and Token-Critic in their sampling processes, and propose Enhanced Sampling Scheme (ESS) to overcome these limitations. ESS explicitly ensures both sample diversity and fidelity, and consists of three stages: Naive Iterative Decoding, Critical Reverse Sampling, and Critical Resampling. ESS starts by sampling a token set using the naive iterative decoding as proposed in MaskGIT, ensuring sample diversity. Then, the token set undergoes the critical reverse sampling, masking tokens leading to unrealistic samples. After that, critical resampling reconstructs masked tokens until the final sampling step is reached to ensure high fidelity. Critical resampling uses confidence scores obtained from a self-Token-Critic to better measure the realism of sampled tokens, while critical reverse sampling uses the structure of the quantized latent vector space to discover unrealistic sample paths. We demonstrate significant performance gains of ESS in both unconditional sampling and class-conditional sampling using all the 128 datasets in the UCR Time Series archive.
+
+[[paper]()]
+[[arXiv](https://arxiv.org/abs/2303.04743)]
+[[github](https://github.com/ML4ITS/TimeVQVAE)]

content/publications/pub6.md

@@ -13,3 +13,7 @@ post_meta = ["categories", "translations"]
 
 ## Abstract
 This paper presents a novel sampling scheme for masked non-autoregressive generative modeling. We identify the limitations of TimeVQVAE, MaskGIT, and Token-Critic in their sampling processes, and propose Enhanced Sampling Scheme (ESS) to overcome these limitations. ESS explicitly ensures both sample diversity and fidelity, and consists of three stages: Naive Iterative Decoding, Critical Reverse Sampling, and Critical Resampling. ESS starts by sampling a token set using the naive iterative decoding as proposed in MaskGIT, ensuring sample diversity. Then, the token set undergoes the critical reverse sampling, masking tokens leading to unrealistic samples. After that, critical resampling reconstructs masked tokens until the final sampling step is reached to ensure high fidelity. Critical resampling uses confidence scores obtained from a self-Token-Critic to better measure the realism of sampled tokens, while critical reverse sampling uses the structure of the quantized latent vector space to discover unrealistic sample paths. We demonstrate significant performance gains of ESS in both unconditional sampling and class-conditional sampling using all the 128 datasets in the UCR Time Series archive.
+
+[[paper]()]
+[[arXiv](https://arxiv.org/abs/2204.02697)]
+[[github]()]

content/publications/pub7.md

@@ -13,3 +13,7 @@ post_meta = ["categories", "translations"]
 
 ## Abstract
 Self-supervised learning (SSL) has had great success in both com- puter vision and natural language processing. These approaches often rely on cleverly crafted loss functions and training setups to avoid feature collapse. In this study, the effectiveness of mainstream SSL frameworks from computer vision and some SSL frameworks for time series are evaluated on the UCR, UEA and PTB-XL datasets, and we show that computer vision SSL frameworks can be effective for time series. In addition, we propose a new method that improves on the recently proposed VICReg method. Our method improves on a covariance term proposed in VICReg, and in addition we aug- ment the head of the architecture by an IterNorm layer that accel- erates the convergence of the model.
+
+[[paper]()]
+[[arXiv](https://arxiv.org/abs/2109.00783)]
+[[github]()]

content/publications/pub8.md

@@ -13,3 +13,7 @@ post_meta = ["categories", "translations"]
 
 ## Abstract
 Self-supervised learning for image representations has recently had many breakthroughs with respect to linear evaluation and fine-tuning evaluation. These approaches rely on both cleverly crafted loss functions and training setups to avoid the feature collapse problem. In this paper, we improve on the recently proposed VICReg paper, which introduced a loss function that does not rely on specialized training loops to converge to useful representations. Our method improves on a covariance term proposed in VICReg, and in addition we augment the head of the architecture by an IterNorm layer that greatly accelerates convergence of the model. Our model achieves superior performance on linear evaluation and fine-tuning evaluation on a subset of the UCR time series classification archive and the PTB-XL ECG dataset.
+
+[[paper]()]
+[[arXiv]()]
+[[github]()]

content/publications/pub9.md

@@ -13,3 +13,7 @@ post_meta = ["categories", "translations"]
 
 ## Abstract
 One of the challenges for the diabetic patients is to regulate the amount of glucose in the blood. Early and reliable meal detection represents one relevant issue to develop more effective treatments. This paper presents a comparison of different classifiers for early meal detection using abdominal sounds. The data presented in the paper is obtained from two different equipment and the classifiers are trained and tested on twelve recordings. The results show that neural networks and convolutional neural networks provide better average detection time (2.875 min and 2.791 min, respectively) than alternative methods recently proposed, and no false positives are observed during testing. Early and reliable meal detection eases the mental burden of the diabetic patients from documenting every meal in the controller and also reduces the risk of hypoglycemia.
+
+[[paper](https://ieeexplore.ieee.org/document/9827841/)]
+[[arXiv]()]
+[[github]()]