Originally published on Towards AI the World's Leading AI and Technology News and Media Company. If you are building an AI-related product or service, we invite you to consider becoming an AI sponsor. At Towards AI, we help scale AI and technology startups. Let us help you unleash your technology to the masses. Last month DeepMind published their new NLP model called RETRO (Retrieval-Enhanced TRansfOrmer) which according to the paper, is a leap forward in the NLP world in multiple aspects.

Autodesk's Will Harris, Foundry's Mathieu Mazerolle and Unity Technologies' Brian Gaffney will discuss how their companies are incorporating machine learning into software tools to make higher quality and more realistic visual effects and boost production speed. Visual Effects Supervisor Ryan Laney will describe the novel way artificial intelligence and machine learning were used to mask the identities of interview subjects in the award-winning HBO documentary Welcome to Chechnya. "Machine learning is poised to transform visual effects production, accelerating workflows and paving the way for a new generation of astonishingly real visual effects," says Barry Goch, who will moderate the discussion. "Will Harris, Mathieu Mazerolle and Brian Gaffney will demonstrate game-changing technologies. Ryan Laney will share his experience in applying machine learning to a real-world production."

Modern neural networks require long training to reach decent performance on massive datasets. One common approach to speed up training is model parallelization, where large neural networks are split across multiple devices. However, different device placements of the same neural network lead to different training times. Most of the existing device placement solutions treat the problem as sequential decision-making by traversing neural network graphs and assigning their neurons to different devices. This work studies the impact of graph traversal order on device placement. In particular, we empirically study how different graph traversal order leads to different device placement, which in turn affects the training execution time. Our experiment results show that the best graph traversal order depends on the type of neural networks and their computation graphs features. In this work, we also provide recommendations on choosing graph traversal order in device placement for various neural network families to improve the training time in model parallelization.

If you had the chance to read Part 1 of this article, you will remember that Transfer Learning is a machine learning method where the application of knowledge obtained from a model used in one task, can be reused as a foundation point for another task. If you did not get the chance and don't have a great understanding of Transfer Learning, give it a read it will help you understand this article much better. So let's first go through what Image Recognition is. Image Recognition is the task assigned to computer technology to be able to detect and analyse an object or a feature in an image or video. It is the major area where deep neural networks work their magic as they are designed to recognise patterns.

Businesses have yet to see if Artificial Intelligence, commonly known as AI, will help or hurt them in the future, but they'll soon be a reality …

As a child, I would sit on the balcony of our Dhaka apartment overlooking the pond and flip through our two family photo albums. After the Bangladesh liberation war in 1971, film was scarce and our camera had broken. With nowhere to get it repaired or to buy film, we had no more family photos for almost a decade. There are no photos of me until I was 8 years old. The tiny, gemlike black-and-white prints of my parents and older brother were fragments of my history that, as curator Glen Helfand said, "captured a fraction of a second of activity and fueled narratives for generations." These images were absorbed by my soul, stored as evidence of the stories of my family from before my birth, and are now on my kids' iPhones.

Using artificial intelligence and machine learning, LSU professor Hartmut Kaiser is joining up with a team of scientists to help communities near …

In this paper, we create EMIR, the first-ever Music Information Retrieval dataset for Ethiopian music. EMIR is freely available for research purposes and contains 600 sample recordings of Orthodox Tewahedo chants, traditional Azmari songs and contemporary Ethiopian secular music. Each sample is classified by five expert judges into one of four well-known Ethiopian Kinits, Tizita, Bati, Ambassel and Anchihoye. Each Kinit uses its own pentatonic scale and also has its own stylistic characteristics. Thus, Kinit classification needs to combine scale identification with genre recognition. After describing the dataset, we present the Ethio Kinits Model (EKM), based on VGG, for classifying the EMIR clips. In Experiment 1, we investigated whether Filterbank, Mel-spectrogram, Chroma, or Mel-frequency Cepstral coefficient (MFCC) features work best for Kinit classification using EKM. MFCC was found to be superior and was therefore adopted for Experiment 2, where the performance of EKM models using MFCC was compared using three different audio sample lengths. 3s length gave the best results. In Experiment 3, EKM and four existing models were compared on the EMIR dataset: AlexNet, ResNet50, VGG16 and LSTM. EKM was found to have the best accuracy (95.00%) as well as the fastest training time. We hope this work will encourage others to explore Ethiopian music and to experiment with other models for Kinit classification.

Abstract--User preference music transfer (UPMT) is a new problem in music style transfer that can be applied to many scenarios but remains understudied. Transferring an arbitrary song to fit a user's preferences increases musical diversity and Most music style transfer approaches rely on datadriven methods. In general, however, constructing a large training Figure 1: A demonstration of UPMT: Transferring symbolic input music dataset is challenging because users can rarely provide enough of to new symbolic music that fits a user's preferences based on features their favorite songs. To address this problem, this paper proposes of their favorite music. For example, Marino et al. [17] used prior semantic knowledge in the form of knowledge graphs HERE has been recent growth in research around music style transfer, a technique that transfers the style of to improve image classification performance. Donadello et al. one piece of music to another based on different levels of [18] extracted semantic representations in a knowledge base music representations [1]. Music style transfer is considered to enhance the quality of recommender systems. Despite these important because it increases music variety by reproducing advances, the approaches cannot be directly applied to music, existing music in a creative way.

Due to extensive spread of fake news on social and news media it became an emerging research topic now a days that gained attention. In the news media and social media the information is spread highspeed but without accuracy and hence detection mechanism should be able to predict news fast enough to tackle the dissemination of fake news. It has the potential for negative impacts on individuals and society. Therefore, detecting fake news on social media is important and also a technically challenging problem these days. We knew that Machine learning is helpful for building Artificial intelligence systems based on tacit knowledge because it can help us to solve complex problems due to real word data. On the other side we knew that Knowledge engineering is helpful for representing experts knowledge which people aware of that knowledge. Due to this we proposed that integration of Machine learning and knowledge engineering can be helpful in detection of fake news. In this paper we present what is fake news, importance of fake news, overall impact of fake news on different areas, different ways to detect fake news on social media, existing detections algorithms that can help us to overcome the issue, similar application areas and at the end we proposed combination of data driven and engineered knowledge to combat fake news. We studied and compared three different modules text classifiers, stance detection applications and fact checking existing techniques that can help to detect fake news. Furthermore, we investigated the impact of fake news on society. Experimental evaluation of publically available datasets and our proposed fake news detection combination can serve better in detection of fake news.