In recent years, citizen science campaigns have provided a very good platform for widespread data collection. Within the marine domain, jellyfish are among the most commonly deployed species for citizen reporting purposes. The timely validation of submitted jellyfish reports remains challenging, given the sheer volume of reports being submitted and the relative paucity of trained staff familiar with the taxonomic identification of jellyfish. In this work, hundreds of photos that were submitted to the “Spot the Jellyfish” initiative are used to train a group of region-based, convolution neural networks. The main aim is to develop models that can classify, and distinguish between, the five most commonly recorded species of jellyfish within Maltese waters. In particular, images of the Pelagia noctiluca, Cotylorhiza tuberculata, Carybdea marsupialis, Velella velella and salps were considered. The reliability of the digital architecture is quantified through the precision, recall, score, and score metrics. Improvements gained through the applicability of data augmentation and transfer learning techniques, are also discussed. Very promising results, that support upcoming aspirations to embed automated classification methods within online services, including smart phone apps, were obtained. These can reduce, and potentially eliminate, the need for human expert intervention in validating citizen science reports for the five jellyfish species in question, thus providing prompt feedback to the citizen scientist submitting the report.
29 September 2021 in IEEE Signal Processing Letters
Convolutional Neural Networks (CNNs) have achieved impressive results across many super-resolution (SR) and image restoration tasks. While many such networks can upscale low-resolution (LR) images using just the raw pixel-level information, the ill-posed nature of SR can make it difficult to accurately super-resolve an image which has undergone multiple different degradations. Additional information (metadata) describing the degradation process (such as the blur kernel applied, compression level, etc.) can guide networks to super-resolve LR images with higher fidelity to the original source. Previous attempts at informing SR networks with degradation parameters have indeed been able to improve performance in a number of scenarios. However, due to the fully-convolutional nature of many SR networks, most of these metadata fusion methods either require a complete architectural change, or necessitate the addition of significant extra complexity. Thus, these approaches are difficult to introduce into arbitrary SR networks without considerable design alterations. In this paper, we introduce meta-attention, a simple mechanism which allows any SR CNN to exploit the information available in relevant degradation parameters. The mechanism functions by translating the metadata into a channel attention vector, which in turn selectively modulates the network's feature maps. Incorporating meta-attention into SR networks is straightforward, as it requires no specific type of architecture to function correctly. Extensive testing has shown that meta-attention can consistently improve the pixel-level accuracy of state-of-the-art (SOTA) networks when provided with relevant degradation metadata. Despite average memory/runtime overheads of less than ~2.6%/0.025 seconds for the datasets and models considered, meta-attention improves the performance for both PSNR and SSIM; for PSNR, the gain on blurred/downsampled (X4) images is of 0.2969 dB (on average) and 0.3320 dB for SOTA general and face SR models, respectively. The coding framework used for this paper is available at: https://github.com/um-dsrg/Super-Resolution-Meta-Attention-Networks.
We revisit the P ? = NP question through the joint lenses of time–relative description complexity and automated discovery. Our premise is epistemic rather than ontological: even if polynomial–time algorithms for NP–complete problems exist, they may have very high Kolmogorov (description) complexity and thus be undiscoverable by unaided humans. Formal barriers (relativization, natural proofs, and algebrization) already suggest that familiar techniques are insufficient to separate or collapse P and NP. Regardless of the ultimate truth of P ? = NP, we argue that systematic search in high–K code spaces is valuable today: it may yield stronger heuristics, tighter exponential bases, improved approximation schemes, and fixed–parameter runtimes. To make such artifacts scientifically credible, we advocate a certificate–first workflow ([9,78]) that couples (i) polytime–by–construction skeletons with (ii) machine–checkable evidence (e.g., DRAT/FRAT logs; LP/SDP duals) and (iii) non–uniform search distilled into uniform algorithms. We also note empirical motivation from large language models: scaling laws and energy budgets indicate that high capacity often unlocks new emergent behaviors, while internal mappings remain complex and opaque. The overarching message is pragmatic: capacity (high descriptive complexity) plus certification may provide a principled path to better algorithms and clearer limits without presuming a resolution of P ? = NP. This paper is best described as an position/expository essay. We synthesize existing work from complexity theory, Kolmogorov complexity, and algorithmic discovery, and offers a rational justification for a shift in emphasis: from the elusive goal of discovering polynomial-time algorithms for NP-complete problems, to the tractable and fruitful pursuit of discovering high-performance heuristics and approximation methods via automated search and learning.
The celebrated question of whether continues to define the boundary between the feasible and the intractable in computer science. In this paper, we revisit the problem from two complementary angles: Time-Relative Description Complexity and automated discovery, adopting an epistemic rather than ontological perspective. Even if polynomial-time algorithms for NP-complete problems do exist, their minimal descriptions may have very high Kolmogorov complexity. This creates what we call an epistemic barrier, making such algorithms effectively undiscoverable by unaided human reasoning. A series of structural results—relativization, Natural Proofs, and the Probabilistically Checkable Proofs (PCPs) theorem—already indicate that classical proof techniques are unlikely to resolve the question, which motivates a more pragmatic shift in emphasis. We therefore ask a different, more practical question: what can systematic computational search achieve within these limits? We propose a certificate-first workflow for algorithmic discovery, in which candidate algorithms are considered scientifically credible only when accompanied by machine-checkable evidence. Examples include Deletion/Resolution Asymmetric Tautology (DRAT)/Flexible RAT (FRAT) proof logs for SAT, Linear Programming (LP)/Semidefinite Programming (SDP) dual bounds for optimization, and other forms of independently verifiable certificates. Within this framework, high-capacity search and learning systems can explore algorithmic spaces far beyond manual (human) design, yet still produce artifacts that are auditable and reproducible. Empirical motivation comes from large language models and other scalable learning systems, where increasing capacity often yields new emergent behaviors even though internal representations remain opaque. This paper is best described as a position and expository essay that synthesizes insights from complexity theory, Kolmogorov complexity, and automated algorithm discovery, using Time-Relative Description Complexity as an organising lens and outlining a pragmatic research direction grounded in verifiable computation. We argue for a shift in emphasis from the elusive search for polynomial-time solutions to the constructive pursuit of high-performance heuristics and approximation methods grounded in verifiable evidence. The overarching message is that capacity plus certification offers a principled path toward better algorithms and clearer scientific limits without presuming a final resolution of .
To date, the best-performing blind super-resolution (SR) techniques follow one of two paradigms: A) generate and train a standard SR network on synthetic low-resolution - high-resolution (LR - HR) pairs or B) attempt to predict the degradations an LR image has suffered and use these to inform a customised SR network. Despite significant progress, subscribers to the former miss out on useful degradation information that could be used to improve the SR process. On the other hand, followers of the latter rely on weaker SR networks, which are significantly outperformed by the latest architectural advancements. In this work, we present a framework for combining any blind SR prediction mechanism with any deep SR network, using a metadata insertion block to insert prediction vectors into SR network feature maps. Through comprehensive testing, we prove that state-of-the-art contrastive and iterative prediction schemes can be successfully combined with high-performance SR networks such as RCAN and HAN within our framework. We show that our hybrid models consistently achieve stronger SR performance than both their non-blind and blind counterparts. Furthermore, we demonstrate our framework's robustness by predicting degradations and super-resolving images from a complex pipeline of blurring, noise and compression.
The maximum clique problem (MCP) is a fundamental problem in graph theory and in computational complexity. Given a graph G, the problem is that of finding the largest clique (complete subgraph) in G. The MCP has many important applications in different domains and has been much studied. The problem has been shown to be NP-Hard and the corresponding decision problem to be NP-Complete. All exact (optimal) algorithms discovered so far run in exponential time. Various meta-heuristics have been used to approximate the MCP. These include genetic and memetic algorithms, ant colony optimization, greedy algorithms, Tabu algorithms, and simulated annealing. This study presents a critical examination of the effectiveness of applying genetic algorithms (GAs) to the MCP compared to a purely stochastic approach. Our results indicate that Monte Carlo algorithms, which employ random searches to generate and then refine sub-graphs into cliques, often surpass genetic algorithms in both speed and capability, particularly in less dense graphs. This observation challenges the conventional reliance on genetic algorithms, suggesting a reevaluation of the roles of the crossover and mutation operators in exploring the solution space. We observe that, in some of the denser graphs, the recombination strategy of genetic algorithms shows unexpected efficacy, hinting at the untapped potential of genetic methods under specific conditions. This work not only questions established paradigms but also opens avenues for exploring algorithmic efficiency in solving the MCP and other NP-Hard problems, inviting further research into the conditions that favor purely stochastic methods over genetic recombination and vice versa.