Brazil

Query performance prediction (QPP) is the namegiven to approaches that aim at predicting, usually post-query, thequality of results generated by an IR system. Yet, the indirect waythat standard QPP approaches estimate performance is opaque,has little relation to the way users evaluate queries, and islimited in application. An alternative, more challenging way tolook at the QPP problem is as arelevance predictionproblem.Instead of resulting in a quality estimate, relevance predictionaims at predicting the relevance of the top-k results in a rank.When accurate, those predictions can be directly employed as amechanism to improve the initial rank. In this work, we explorea learning framework based on sequence labeling combined withcontextual features for relevance prediction. Sequence labelingis a common approach to deal with sequential data, learningfrom the sequence to generate a sequence of predicted labels.The context within a rank, that is, the neighborhood of rankedobjects, has been successfully explored within rank-aggregationand re-ranking scenarios, and here we extend the idea to devisean effective relevance prediction feature. We show that thecombination of both can beat the baseline relevance predictionapproach, performing effective prediction, that is accurate for upto k = 30 positions.

IWSSIP 2020

Contextual Features and Sequence Labeling Techniques for Relevance Prediction in Retrieval

infor-mation retrieval

query perf. prediction

relevance prediction

machine learning

sequence labeling

**Welcome to EMNLP 2023!**

On behalf of the EMNLP 2023 Organizing Committee, I extend a warm and heartfelt welcome to all of you to the 2023 Conference on Empirical Methods in Natural Language Processing (EMNLP). It is with immense pleasure and excitement that we gather here in Singapore, a vibrant hub of innovation and technological advancement.

The conference program is packed with insightful presentations, thought-provoking workshops, and engaging networking opportunities. In addition to the technical sessions, we have also planned several social events that will provide you with the opportunity to connect with your colleagues.

The conference is held in person at the Resorts World Convention Centre in Singapore, and available on-line with the help of Underline.

For those in Singapore, I hope that you will find time to explore the exciting Sentosa Island, Gardens by the Bay, Singapore Botanic Gardens and the many other attractions unique to Singapore.

Of course, we are grateful to our sponsors and partners for their generous support of EMNLP 2023, whose contributions make it possible for us to host this world-class event.

Yuji Matsumoto (RIKEN AIP) 
EMNLP 2023 General Chair

To access the **EMNLP 2023** event page on Underline, you need to register for the Conference. 
Please follow **[this link](https://2023.emnlp.org/registration/)** for more details.

EMNLP 2023

EMNLP 2023 took place in Singapore from Dec 6th to Dec 10th, 2023.

technical paper

The 27th International Conference on Systems, Signals and Image Processing.

Personalized music systems usually rely on manual song annotations (tags) as a mechanism for querying and navigating large music collections. However, the manual annotation is a hard task given the large amount of music available nowadays. Automatic song annotation based on content analysis is a potential solution to this problem and has recently been gaining attention. In this work, we propose to extend the Stacking prediction framework to use Convolutional Neural Networks (CNNs) in order to improve music tag annotation task. In general, the Stacked prediction consists of a technique in which the output of the first stage of learning is used as input in the second stage. In our work we have tried two proposals of extension. The first one consists of using the weights learned by the CNN in the first stage of training as input in the second stage. In the second proposal, we use the autoencoder technique with the weights learned by CNN in the first stage to generate images 50% smaller than the ones used as input of the first stage. The weights and the images obtained in the first stage are used as input in the second stage. We evaluated our proposals with five different CNN models in three datasets well-known in the literature (FMA, MillionSong, and MagnaTagATune), obtaining interesting results.

Improving Musical Tag Annotation with Stacking and Convolutional Neural Networks

The number of Distributed Denial of Service (DDoS) attacks using IoT devices has increased in recent years. Reasons for this growth include the security limitations of IoT devices, the number of devices, and their geographic distribution. Developing mechanisms to detect and mitigate DDoS attacks in this scenario is a current challenge in the area of network security. In the literature review, it is seen that recent academic works still tries to find the best way to combat this type of threat, with proposals that need to be tested against modern datasets that contain a diversity of modern denial of service attacks. This work proposes a detection module for an IoT controller that uses Machine Learning techniques to classify network traffic. The system was designed in the Software-Defined Networks (SDN) context and evaluated on an emulated platform using three actual and well-know datasets present in the literature. The results, at a sampling rate of 20% of network traffic, show a high precision, above 93%, a low false alarm rate, and detection rate of attacks above 96%, using a low profile emulated device.

Smart Detection-IoT: A DDoS Sensor System for Internet of Things

Ultrasonic signals are widely used to detect defectsand characterize materials. In this field, the Total Focusing Method (TFM) imaging approach is commonly applied to post-process the ultrasonic data acquired by the Full Matrix Capture (FMC) of an array of ultrasonic transducers. Although the traditional FMC-TFM technique is considered as a reference standard, the energy transmitted in the material is limited by the use of a single transmitting element of the array and hence may provide a poor signal-to-noise ratio of the reconstructed image. To overcome this limitation and improve the image quality of highly noisy materials, this paper combines two noise-reduction methods, namely a modified Spatially Averaged Sub-ApertureCorrelation Imaging and virtual sources. In the former, we further propose a modification in order to avoid the manual fine-tuning of a parameter presented in the original approach. The second consists of using several transmitting elements to increase theenergy transmitted in the medium. Results showed that this combination was able to increase in 39dB the signal-to-noise ratio. This result is 4.2dB above the sum of individual gains of each technique which indicates a synergistic effect on increasing the image quality. Moreover, the use of virtual sources was capable of improving the reconstructed image where initially it is not possible to distinguish the noise from the desired signal. The application of the correlation approach in this scenario increased the quality of the reconstructed image significantly. The proposed combination was validated in a finite element model of a steel block with side-drilled holes used to mimic internal flaws in the material. Several noise levels were added to the numerically simulated signals in order to evaluate the performance of the proposed approach.

Improvement of Ultrasonic Image by a Combination of Correlation-Based Image Processing and Virtual Sources Applied to Non-Destructive Evaluation

In this contribution a design of a full polarimetric multi-channel Ground Penetrating Radar (GPR) with optimized antenna switching is presented. Firstly, a single-polarized system was used to obtain multiple measurements of a static scene with different antenna orientation and providing a full polarimetric dataset. Laboratory experiments have been performed on metal test landmines and where promising results have been obtained. The single-polarized radar system was modified to a multichannel system with focus on decreasing the scanning time. The selected frequency range covers the band of 550 MHz to 2.7 GHz, which corresponds in soil to a range resolution of approximately 7 cm. There was no need to use a Power Amplifier (PA) at the transmitter stage, since a highly sensitive receiver is implemented. This greatly decrease the power consumption and makes the device suitable for battery powered platforms. Physical contact between the ground and antennas was avoided with using the air-launched measurement technique, where additionally the level of safety is higher.

Design of a Full Polarimetric GPR System for Landmine Detection

The presentation focus signal processing techiniques appliend on the ATLAS experiment, present on the LHC machine. It compares two energy estimation methods in high pile-up scenario using LHC collision data.

Energy Reconstruction Performance in the ATLAS Tile Calorimeter Operating at High Event Rate Conditions Using LHC Collision Data

Blur and other undesirable visual artifacts are often produced from simple interpolation methods when applied to magnification from low-resolution into higher resolution images. This work evaluates the effect of existing sharpening filters when applied to images obtained from interpolation in order to determine whether image quality can be improved. Additionally, a novel sharpening filter is proposed which uses information from the interpolation process to prevent the occurrence of blur. The novel filter is shown to achieve better results when compared to other sharpening filters from the literature.

An Empirical Evaluation of Sharpening Filters Applied to Magnified Images

In the new context of smart grids (SGs), new types of electric disturbances may occur in the electrical power systems (EPSs) current and voltage waveforms. The detection and the study of such disturbances can be useful in the operation and control of the EPSs. In order to identify any waveform abnormality or variation, novelty detectors might be used. This work presents a novelty detection technique for power quality signals based on surrogates and distances calculated in the time-frequency domain. Surrogates are signals created with the same magnitude of the Fourier spectrum from the original signal, but with the phase spectrum generated from a random uniform distribution. Tests using several cases of the most common power quality disturbances using synthetic signals were conducted. A comparison with an energy-based novelty detector currently present in the literature is performed, showing better performance results in terms of the receiver operating characteristic (ROC) curve parameters. Also, a real-life waveform signal is used to show the presented technique in action. The surrogate technique presented in this work can be useful in the definition of the novelties ground-truth for real power signal waveforms.

Novelty Detection in Power Quality Signals with Surrogates: a Time-Frequency Technique

One of the goals of the music information retrieval (MIR) community is to research new methods and create new systems that can efficiently and effectively retrieve and recommend songs from large databases of music content. Despite the volume of research in the area, there is a lack of music databases to support these works, i.e. databases that comply with some quite desirable requirements for the development of researches, such as: a huge amount of music pieces, the audio signal availability and a great diversity of audio attributes. In order to contribute to the MIR community, we present Music4All, a new music database which contains metadata, tags, genre information, 30-seconds audio clips, lyrics, and so on. Additionally, we also exemplify some MIR tasks that may benefit from our database and compare it with other databases proposed in the literature. To request the Music4All database, send an e-mail to contact4music4all@gmail.com.

Music4All: A New Music Database and its Applications

In this paper, we propose a method for face recognition
by applying data augmentation and transfer learning in pretrained Convolutional Neural Networks (CNNs). Our main focus
is to analyze the power of data augmentation for face recognition
systems with CNN transfer learning. We extracted features from
the images and trained a KNN classifier using the VGG-Face
CNN.

Data Augmentation for Face Recognition with CNN Transfer Learning

Around the world, there are many people with disabilities; it is estimated that 39 million people are blind and 246 million have limited vision, giving a total of 285 million visually impaired people. The use of information and communication technologies can help disabled people to achieve greater independence, quality of life and inclusion in social activities by increasing, maintaining or improving their functional capacities. In this context, this paper presents an automatic methodology for identifying banknotes that can be widely used by people with visual impairment. For this, we evaluated a set of four point of interest detectors, two descriptors, seven ways of generating the image signature, and six classification methodologies, which can be used as a basis for the development of applications for the identification of banknotes. Experiments performed on US Dollar (USD), Euro (EUR) and Brazilian Real Banknotes (BRL) obtained rates of accuracy of 99.78%, 99.12%, and 96.92%, respectively.

Premium content

Downloads

Next from IWSSIP 2020

Improving Musical Tag Annotation with Stacking and Convolutional Neural Networks

Similar lecture

Enhancing Generative Retrieval with Reinforcement Learning from Relevance Feedback | VIDEO

Stay up to date with the latest Underline news!

PRESENTATIONS

CONFERENCES

COMPANY

RESOURCES