About
In this post, I will talk about my experience in attending the International Joint Conference on Neural Networks (IJCNN) 2021 organized by the International Neural Network Society. It is a very renowned and popular conference that encourages papers on neural networks theory, analysis and applications covering various research areas like bioinformatics, data mining, deep learning, sensor networks, machine learning, neuroengineering, etc. This conference was held online on a virtual platform. The complete proceeding was made available to all the participants via a single downloadable link. We have our paper entitled “Fruit classification using deep feature maps in the presence of deceptive similar classes” accepted for publication at the conference. Following this, I participated as a presenter of our paper. All participants received a comprehensive informational email with access credentials and instructions for how to access all content in the virtual platform of the conference.
Day 1
Attended welcome and plenary talks, and paper presentations in various tracks like Adversarial Machine Learning and Cyber Security, Artificial Intelligence and Security (AISE), Healthcare Analytics: Improving Healthcare outcomes using Multimedia Data Analytics and Neural Network Models on the first day of the conference. The paper and poster presentations were shown using pre-recorded videos of 15 min. followed by 5 min. Q&A in several parallel sessions or rooms. Also, I liked the plenary talk by Prof. Marios M. Polycarpou on Smart Interactive Buildings covering distributed and smart Heating, Ventilation, and Air Conditioning (HVAC) systems and other monitoring systems. It was said that the selection of appropriate control objective is a very crucial step and it is usually taken as a given for the automated system. There are some methods available based on semantics to adjust the control objectives based on the requirements of the humans. In another plenary talk, Prof. Karl J. Friston talked about active interfaces covering decision making and behaviour choice that are governed by value functions and reward. Overall, the platform was well managed following the excellent presentations and schedule. However, the only thing that lacks is the participation of the attendees in asking questions, mostly it is the host who asks questions which I feel should be improved from the participation side.
Day 2
Day 2 began with the paper presentations where various tracks are covered in parallel sessions: Bayesian Neural Networks & AutoML applications, Current Trend of Machine Learning in Computer Vision, Learning from Imbalanced and Difficult Data, and Special Session on Federated Learning and Cooperative Neural Networks, and other topics. Later, a keynote talk on the interesting topic “What neuroimaging can tell about human brain function, Riitta Salmelin” is delivered by Prof. Riitta Salmelin. The talk filled us with the state-of-the-art developments in the neuroimaging domain and what is the upcoming phase of developments and advancements to further advance from group-level descriptions to quantitative model-based individual-level predictions. Later, after the keynote talk again different paper tracks are presented. I also presented our accepted paper entitled “Fruit classification using deep feature maps in the presence of deceptive similar classes” authored by Mohit Dandekar, Narinder Singh Punn, Sanjay Kumar Sonbharda and Sonali Agarwal under the track “Deep neural networks VI” ( the paper is also available here). It was a good experience where I interacted with other participants and session chairs.
Day 3
This day initiated with the plenary talk by Prof. Peter Tino on “Fascinating World of Recurrent Networks: A Personal View”. I liked his interesting talk where he talked about the learning process from the dynamic data, where the order in which the data is received does impact the performance of the proposed solution (insight of temporal structure of data). Following this, various deep learning-based models are discussed covering state-of-the-art recurrent neural networks (RNN) with deep insight into its working mechanism. Later, in another plenary talk on “On Presuppositions of Machine Learning: A Best-fitting Theory” by Prof. Zongben Xu, we were presented with different machine learning hypothesis approaches that follow the proposed solution. Afterwards, the paper and poster presentations continued on various tracks in parallel sessions covering machine learning and deep learning, data analytics and computation intelligence, supervised and unsupervised learning.
Day 4
On day 4 more papers are covered for various tracks such as bio-inspired systems and applications, deep learning, semi-supervised learning, ensemble modelling, etc. I attended 5-8 presentations, where I liked the paper presentations: “PANDA : Perceptually Aware Neural Detection of Anomalies”, “Unsupervised Post-Tuning of Deep Neural Networks” and “A Meta-Learning Approach for Automated Hyperparameter Tuning in Evolving Data Streams”. Later, a keynote talk was delivered on Self-Organized Criticality (SOC) in the Brain by Prof. Dietmar Plenz, where SOC can be expressed as the evolution capability of complex systems into 2\sup{nd}-order phase transition. At this phase interactions between systems, components lead to scale-invariant events that are beneficial for system performance. This plenary talk is followed by paper presentations on different applications of neural networks, neuroengineering and perception.
Day 5
The day started with an interesting keynote talk by Prof. Nikola Kasabov on “Transfer Learning and Knowledge Transfer Between Humans and Machines with Brain-Inspired Spiking Neural Networks for Adaptable and Explainable AI”. In this talk, a brain-inspired spiking neural network architectures are discussed that uses transfer learning to adapt new tasks/classes/categories incrementally by using self-organizing learning principles. The later session continued with the paper presentations covering various tracks.
Final Day
On the day 6 various tutorial sessions are covered covering Accelerating Deep Learning Computation, Deep Learning for Graphs, Machine Learning for Brain-Computer Interfaces, Deep learning applied to the viral genome classification, Randomization Based Deep and Shallow Learning Methods for Classification and Forecasting, etc. I happened to link Accelerating Deep Learning Computation and Deep Learning for Graphs tutorials, since it is my field of interest. These tutorials also followed some paper presentations, workshops and competitions. Later the day and conference ended with the closing remarks. I seek forward to participate in this conference again.