Scalable Vector Graphics (SVG) offer a powerful format for representing visual designs as interpretable code. Recent advances in vision-language models (VLMs) have enabled high-quality SVG generation by framing the problem as a code generation task and leveraging large-scale pretraining. VLMs are particularly suitable for this task as they capture both global semantics and fine-grained visual patterns, while transferring knowledge across vision, natural language, and code domains. However, existing VLM approaches often struggle to produce faithful and efficient SVGs because they never observe the rendered images during training. Although differentiable rendering for autoregressive SVG code generation remains unavailable, rendered outputs can still be compared to original inputs, enabling evaluative feedback suitable for reinforcement learning (RL). We introduce Reinforcement Learning from Rendering Feedback, an RL method that enhances SVG generation in autoregressive VLMs by leveraging feedback from rendered SVG outputs. Given an input image, the model generates SVG roll-outs that are rendered and compared to the original image to compute a reward. This visual fidelity feedback guides the model toward producing more accurate, efficient, and semantically coherent SVGs. \method significantly outperforms supervised fine-tuning, addressing common failure modes and enabling precise, high-quality SVG generation with strong structural understanding and generalization.

Unsupervised object discovery is becoming an essential line of research for tackling recognition problems that require decomposing an image into entities, such as semantic segmentation and object detection. Recently, object-centric methods that leverage self-supervision have gained popularity, due to their simplicity and adaptability to different settings and conditions. However, those methods do not exploit effective techniques already employed in modern self-supervised approaches. In this work, we consider an object-centric approach in which DINO ViT features are reconstructed via a set of queried representations called slots. Based on that, we propose a masking scheme on input features that selectively disregards the background regions, inducing our model to focus more on salient objects during the reconstruction phase. Moreover, we extend the slot attention to a multi-query approach, allowing the model to learn multiple sets of slots, producing more stable masks. During training, these multiple sets of slots are learned independently while, at test time, these sets are merged through Hungarian matching to obtain the final slots. Our experimental results and ablations on the PASCAL-VOC 2012 dataset show the importance of each component and highlight how their combination consistently improves object localization. Our source code is available at: github.com/rishavpramanik/maskedmultiqueryslot

In today’s era of data-driven digital society, there is a huge demand for optimized solutions that essentially reduce the cost of operation, thereby aiming to increase productivity. Processing a huge amount of data, like the Microarray based gene expression data, using machine learning and data mining algorithms has certain limitations in terms of memory and time requirements. This would be more concerning, when a dataset comes with redundant and non-important information. For example, many report-based medical datasets have several non-informative attributes which mislead the classification algorithms. To this end, researchers have been developing several feature selection algorithms that try to discard the redundant information from the raw datasets before feeding them to machine learning algorithms. Metaheuristic based optimization algorithms provide an excellent option to solve feature selection problems. In this paper, we propose a music-inspired harmony search (HS) algorithm based wrapper feature selection method. At the beginning, we use a chaotic mapping to initialize the population of the HS algorithm in order to better coverage of the search space. Further to complement the inferior exploitation of the HS algorithm, we integrate it with the Late Acceptance Hill Climbing (LAHC) method. Thus the combination of these two algorithms provides a good balance between the exploration and exploitation of the HS algorithm. We evaluate the proposed feature selection method on 15 UCI datasets and the obtained results are found to be better than many state-of-the-art methods both in terms of the classification accuracy and the number of features selected. To evaluate the effectiveness of our algorithm, we utilize a combination of precision, recall, F1 score, fitness value, and execution time as performance indicators. These metrics enable us to obtain a comprehensive assessment of the algorithm’s abilities and limitations. We also apply our method on 3 microarray based gene expression datasets used for prediction of cancer to ensure the scalability and robustness as a feature selection method in real-life scenarios. In addition to this, we test our approach using the COVID-19 dataset, and it performs better than several metaheuristic based optimization techniques.