Rakesh Das

Asmit Bandyopadhyay, Anindita Das Bhattacharjee, Rakesh Das

Hyperspectral image (HSI) classification faces critical challenges, including high spectral dimensionality, complex spectral-spatial correlations, and limited training samples with severe class imbalance. While CNNs excel at local feature extraction and transformers capture long-range dependencies, their isolated application yields suboptimal results due to quadratic complexity and insufficient inductive biases. We propose CLAReSNet (Convolutional Latent Attention Residual Spectral Network), a hybrid architecture that integrates multi-scale convolutional extraction with transformer-style attention via an adaptive latent bottleneck. The model employs a multi-scale convolutional stem with deep residual blocks and an enhanced Convolutional Block Attention Module for hierarchical spatial features, followed by spectral encoder layers combining bidirectional RNNs (LSTM/GRU) with Multi-Scale Spectral Latent Attention (MSLA). MSLA reduces complexity from $\mathcal{O}(T^2D)$ to $\mathcal{O}(T\log(T)D)$ by adaptive latent token allocation (8-64 tokens) that scales logarithmically with the sequence length. Hierarchical cross-attention fusion dynamically aggregates multi-level representations for robust classification. Experiments conducted on the Indian Pines and Salinas datasets show state-of-the-art performance, achieving overall accuracies of 99.71% and 99.96%, significantly surpassing HybridSN, SSRN, and SpectralFormer. The learned embeddings exhibit superior inter-class separability and compact intra-class clustering, validating CLAReSNet's effectiveness under limited samples and severe class imbalance.

Anushko Chattopadhyay, Ambuj, Rakesh Das et al.

We introduce conclusive identification task for classical channels: a receiver identifies transmitted inputs without error when possible, and responds inconclusively when outputs are ambiguous. For a symmetric not-fully-corrupted channel $N : X \to X$, the single-shot conclusive identification index $\mathrm{ci}_\circ(N)$ counts the maximum number of conclusively identifiable inputs. We show $\mathrm{ci}_\circ(N)$ exhibits a striking superactivation phenomenon: a channel with $\mathrm{ci}_\circ(N) = 0$ achieves $\mathrm{ci}_\circ(N \otimes \mathrm{id}^c_β) = |X|$ when assisted by a perfect classical channel of dimension $β< |X|$. The minimum classical assistance required equals the chromatic number $χ(\mathtt{S}_N)$ of the channel's support graph $\mathtt{S}_N$. We provide channel families where the superactivation gap $\mathrm{ci}_\circ(N \otimes \mathrm{id}^c_β) - \mathrm{ci}_\circ(\mathrm{id}^c_β)$ can be made arbitrarily large. A noiseless quantum channel of dimension equal to the orthogonal rank $ξ(\mathtt{S}_N)$ suffices, yielding a strict quantum advantage whenever $ξ(\mathtt{S}_N) < χ(\mathtt{S}_N)$. This advantage is demonstrated through three explicit constructions motivated by combinatorial and algebraic state-independent, and state-dependent proofs of Kochen-Specker contextuality. Via the co-normal product of graphs, we analyze the scaling of the quantum advantage ratio $χ_f(\mathtt{S}_N)/ξ(\mathtt{S}_N)$, and present a channel for which quantum assistance is exponentially more efficient than classical. Our results establish $\mathtt{S}_N$, rather than the confusability graph $\mathtt{G}_N$, as the natural combinatorial object for conclusive identification, revealing that channels deemed useless under Shannon's zero-error framework can exhibit rich superactivation and quantum advantage, with deep connections to quantum contextuality.