ISIT 2021
215 papers
The 2021 IEEE International Symposium on Information Theory (ISIT) featured 246 papers addressing core challenges in communication and data processing. A primary research direction involved the intersection of information theory and machine learning, with significant focus on information-theoretic generalization, the bottleneck principle, and high probability generalization bounds. Other prominent themes included semantic communication for wireless image transmission, unsourced random access for massive machine-type communications, and the development of provable guarantees for algorithmic performance.
Technical contributions also explored bridging the gap between combinatorial existence and efficient decoding, alongside advancements in parallel architectures for sequential decoding. Notable papers investigated topics such as sparse recovery with shuffled labels, analog joint source-channel coding using deep neural networks, and new uncertainty measures like $t$-entropy. Prolific authors during this edition included Flávio P. Calmon, Antonia Wachter-Zeh, Yury Polyanskiy, and Deniz Gunduz. The year’s research reflected a trend toward integrating deep learning with classical information theory, evidenced by work on distributed interference alignment and Riemannian manifold optimization for learning orthogonal representations.
Top Research Directions
Research directions most represented by ISIT 2021 papers.
Top Authors
Impactful Papers
Papers from ISIT 2021 that shaped research clusters.
Influential
Sketching and Sequence Alignment: A Rate-Distortion Perspective
influential
Overcoming the Quadratic Bottleneck in S
Mixed Dimension Embeddings with Application to Memory-Efficient Recommendation Systems
influential
Frequency Aware Embedding Allocation
Generalized Spatially Coupled Parallel Concatenated Convolutional Codes With Partial Repetition
influential
Inherent Rate Adaptation via Rateless Co
Learning under Distribution Mismatch and Model Misspecification
influential
Quantifying Generalization via Informati
Regret Bounds for Safe Gaussian Process Bandit Optimization
influential
Safe Exploration and Optimization with G
Linear Coding for AWGN Channels with Noisy Output Feedback via Dynamic Programming
influential
Overcoming the Noiseless Feedback Assump
Fast-Rate Loss Bounds via Conditional Information Measures with Applications to Neural Networks
influential
Fast Generalization Rates for Interpolat
Numerically stable coded matrix computations via circulant and rotation matrix embeddings
influential
Algebraic Redundancy for Straggler Resil