| Workshop 5 |
| Title of workshop: Deep Learning Applications in Astronomy |
| Details: We propose a workshop on applications of Deep Learning to astronomy. Machine Learning, particularly using Artificial Neural Networks (ANN), Support Vector Machines (SVM) and other related tools, has been extensively applied to regression and classification problems in astronomy. These include star-galaxy classification when the galaxies are so distant that they have near-point like images, stellar classification using their spectra, distinguishing between star and quasars on the basis of their colours, morphological classification of galaxies and the estimation of the redshift of extragalactic objects using photometric data. In the traditional applications of machine learning, when input images or spectra are used, representative features of the concerned objects are used to train an ANN, say, to be able to distinguish between different classes, like stars and galaxies. The trained network can then be used to classify large samples of objects without human intervention. Deep Learning algorithms use many layers of nonlinear processing units, providing much greater depth as well as width compared to traditional machine learning techniques. There is therefore a very large number of adjustable weights, so that whole images or high resolution spectra can be directly used as inputs for training, leading to the possibilities of subtle image recognition and classification. In recent times, Deep Learning techniques are being used in many areas where complex structures are to be discovered in high-dimensional data, including image and speech recognition, analysing particle accelerator data and predicting the activity of potential drug molecules and natural language processing. It is also used in economics and finance. In all such applications, Deep Learning surpasses machine learning in the accuracy of results obtained, and can be used to solve problems that were considered to be intractable for machine learning. There is rapid evolution taking place in the theory and techniques of Deep Learning and its applications. The use of Deep Learning in astronomy is just beginning. It has been applied, for example to determining galaxy morphology and classification of stellar spectra. Our group is using Deep Learning to discover bars in galaxies and to identify rare stellar objects from their spectra. The applications to astronomy are expected to grow enormously over the coming years, particularly in the context of large surveys and data sets. A workshop on Deep Learning at the ASI Meeting is therefore very timely: it will serve to introduce young as well as experienced researchers to the emerging techniques and their possible applications to astronomy, which could later be studied in much greater detail for specific cases. The proposers are engaged in several projects using Deep Learning in astronomy as well as other fields and are therefore well placed to conduct a workshop on the topic. |
Topics to be Covered: The programme will consist of a number of talks to be followed by the demos and hands on sessions. The topics to be covered in the talk will include:
The astronomical applications will cover work by the proposers as well as other groups. The lectures will begin at 10 am (earlier if possible) and will continue for about three hours. The afternoon will be devoted to demos and hands-on sessions. The lectures and practical sessions will be covered by the proposers and possibly a few other experts from the country and abroad. |
| Requirements: The level will be suitable for young researchers working in astronomy having familiarity with the use of software packages and the Python programming language. The participants will need to bring a laptop for hands-on work. They will be advised on the software packages they will have to load on their laptops prior to the workshop. Good audio-visual facilities, Wifi with sufficient bandwidth and a number of power outlets will be required for conducting the workshop. |
| Expected Participants: 50 |
| Details on Organisation: Organisers: Ajit Kembhavi (IUCAA), Ninan Sajeeth Philip (St. Thomas College) Ashish Mahabal (Caltech), Sheelu Abraham (IUCAA), Kaustubh Vaghmare(IUCAA) |