{"data":{"featured":{"edges":[{"node":{"frontmatter":{"title":"SCALe: Supervised Contrastive approach for Active Learning","cover":{"childImageSharp":{"gatsbyImageData":{"layout":"constrained","placeholder":{"fallback":"data:image/png;base64,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"},"images":{"fallback":{"src":"/static/5bb8e7db034b836e5b6a447e31a53477/aa288/scale4.png","srcSet":"/static/5bb8e7db034b836e5b6a447e31a53477/b4165/scale4.png 175w,\n/static/5bb8e7db034b836e5b6a447e31a53477/931c2/scale4.png 350w,\n/static/5bb8e7db034b836e5b6a447e31a53477/aa288/scale4.png 700w","sizes":"(min-width: 700px) 700px, 100vw"},"sources":[{"srcSet":"/static/5bb8e7db034b836e5b6a447e31a53477/1e587/scale4.avif 175w,\n/static/5bb8e7db034b836e5b6a447e31a53477/e0928/scale4.avif 350w,\n/static/5bb8e7db034b836e5b6a447e31a53477/d4ce7/scale4.avif 700w","type":"image/avif","sizes":"(min-width: 700px) 700px, 100vw"},{"srcSet":"/static/5bb8e7db034b836e5b6a447e31a53477/03bd4/scale4.webp 175w,\n/static/5bb8e7db034b836e5b6a447e31a53477/38b5a/scale4.webp 350w,\n/static/5bb8e7db034b836e5b6a447e31a53477/d947e/scale4.webp 700w","type":"image/webp","sizes":"(min-width: 700px) 700px, 100vw"}]},"width":700,"height":480}}},"tech":["Python","NLP","SupCL","MLM"],"github":"https://github.com/kaushalrai7797/SCALe-Supervised-Contrastive-approach-for-Active-Learning","external":"https://github.com/kaushalrai7797/SCALe-Supervised-Contrastive-approach-for-Active-Learning/blob/main/SCALe_Supervised_Contrastive_approach_for_Active_Learning.pdf"},"html":"\n- Proposed a novel loss function (supervised contrastive loss) for active learning that generates multiple positives using dropout maskings.
\n- Concluded that using MLM to predict missing tokens doesn't solve the objective of predicting dicriminative sentence embeddings for text classification.
\n- Showed that using contrastive learning is better than MLM for sentence representation in low-data regime. Embeddings generated by contrastive learning are discriminative.
\n- The proposed loss function lead to an improvement of 10% on the Trec-6 dataset and up to 4% on SST-2 dataset on 15% acquired data, thereby beating the current state-of-the-art method.
\n
"}},{"node":{"frontmatter":{"title":"Context Associated Object Removal","cover":{"childImageSharp":{"gatsbyImageData":{"layout":"constrained","placeholder":{"fallback":"data:image/png;base64,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"},"images":{"fallback":{"src":"/static/a836543808b6a28fe79c04f33035c170/bc3b9/remover.png","srcSet":"/static/a836543808b6a28fe79c04f33035c170/a8dfe/remover.png 175w,\n/static/a836543808b6a28fe79c04f33035c170/a3fa1/remover.png 350w,\n/static/a836543808b6a28fe79c04f33035c170/bc3b9/remover.png 700w,\n/static/a836543808b6a28fe79c04f33035c170/87926/remover.png 1400w","sizes":"(min-width: 700px) 700px, 100vw"},"sources":[{"srcSet":"/static/a836543808b6a28fe79c04f33035c170/d6c11/remover.avif 175w,\n/static/a836543808b6a28fe79c04f33035c170/d7667/remover.avif 350w,\n/static/a836543808b6a28fe79c04f33035c170/7ec1a/remover.avif 700w,\n/static/a836543808b6a28fe79c04f33035c170/71a30/remover.avif 1400w","type":"image/avif","sizes":"(min-width: 700px) 700px, 100vw"},{"srcSet":"/static/a836543808b6a28fe79c04f33035c170/307e4/remover.webp 175w,\n/static/a836543808b6a28fe79c04f33035c170/26a00/remover.webp 350w,\n/static/a836543808b6a28fe79c04f33035c170/f23f0/remover.webp 700w,\n/static/a836543808b6a28fe79c04f33035c170/2c2d0/remover.webp 1400w","type":"image/webp","sizes":"(min-width: 700px) 700px, 100vw"}]},"width":700,"height":394}}},"tech":["Python","DeepFillv2","SuperPixel","CV"],"github":null,"external":"https://pdfhost.io/v/o6PR13dc0_Context_associated_Object_Removal_from_images"},"html":"\n- Developed a framework to generate coherent outputs by removing the context (shadow) associated with the object.
\n- Conducted in-depth analysis to conclude that LISA coupled with DeepFillv2 shows the best result.
\n
"}},{"node":{"frontmatter":{"title":"Detecting Hate Induced Memes","cover":{"childImageSharp":{"gatsbyImageData":{"layout":"constrained","placeholder":{"fallback":"data:image/png;base64,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"},"images":{"fallback":{"src":"/static/371ed85288dc304b25cc25c2dc912074/76f9a/hateInducedmeme.png","srcSet":"/static/371ed85288dc304b25cc25c2dc912074/9051f/hateInducedmeme.png 175w,\n/static/371ed85288dc304b25cc25c2dc912074/b2c0b/hateInducedmeme.png 350w,\n/static/371ed85288dc304b25cc25c2dc912074/76f9a/hateInducedmeme.png 700w,\n/static/371ed85288dc304b25cc25c2dc912074/36422/hateInducedmeme.png 1400w","sizes":"(min-width: 700px) 700px, 100vw"},"sources":[{"srcSet":"/static/371ed85288dc304b25cc25c2dc912074/2f9bd/hateInducedmeme.avif 175w,\n/static/371ed85288dc304b25cc25c2dc912074/56394/hateInducedmeme.avif 350w,\n/static/371ed85288dc304b25cc25c2dc912074/565bd/hateInducedmeme.avif 700w,\n/static/371ed85288dc304b25cc25c2dc912074/c4e48/hateInducedmeme.avif 1400w","type":"image/avif","sizes":"(min-width: 700px) 700px, 100vw"},{"srcSet":"/static/371ed85288dc304b25cc25c2dc912074/9b040/hateInducedmeme.webp 175w,\n/static/371ed85288dc304b25cc25c2dc912074/18cc4/hateInducedmeme.webp 350w,\n/static/371ed85288dc304b25cc25c2dc912074/efdb4/hateInducedmeme.webp 700w,\n/static/371ed85288dc304b25cc25c2dc912074/1f42c/hateInducedmeme.webp 1400w","type":"image/webp","sizes":"(min-width: 700px) 700px, 100vw"}]},"width":700,"height":372}}},"tech":["Python","Machine Learning","LSTM","CNN"],"github":"https://github.com/kaushalrai7797/Detecting-Hate-Inducing-Memes-in-Code-Switched-Language","external":"https://github.com/kaushalrai7797/Detecting-Hate-Inducing-Memes-in-Code-Switched-Language/blob/master/Hate_me_not_paper.pdf"},"html":"\n- Introduced a novel triply annotated Indian political Memes (IPM) dataset, which comprises memes from various Indian political events that have taken place post-independence.
\n- Proposed and implemented a binary-channeled CNN and LSTM based model wherein the images are individually processed using the CNN model and the text extracted from the images using the LSTM model. This model gives state-of-the-art results in the domain of offensive image (Memes) classification in code-switched languages.
\n
"}},{"node":{"frontmatter":{"title":"Code Grading and Analysis","cover":{"childImageSharp":{"gatsbyImageData":{"layout":"constrained","placeholder":{"fallback":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAABQAAAANCAYAAACpUE5eAAAACXBIWXMAABYlAAAWJQFJUiTwAAABaElEQVQ4y4WTi46CQAxF+f+vU0KMEEPiAxTFt4CC0M1p0s0M6+6STAqd29vbB4F8eIZhUNv3vdxuN9nv97JareR6vUrTNB5m/AR/EWKrqpLZbCaHw0He77e8Xq//CbkcH/OjaD6fS57nSti2rfwWwwk+ZXJLRlGWZbJcLuX5fHqEY2JPoUvmEtZ1reoej4cqhHSMc2OVsCxLLQsVXdd5AAgo+3K5qIXUffDvdjs5nU4aqyXjTNNUiqLQElFGIHeUyD1l3+939XFvGIjYADBgVSHrsFgsNICpcsEhI99MmNUBR0K7Rz2+zWajsarQ+mREOPHZO4d3MKYKH9aGxqEd2MBKsJ1D/vl81gAGQUk2XXqFGksA9ng8frfnx9pQGkvMkMi23W61FQRT2mQykel0qt+0IIoiWa/X3rIH7vgBJUmiClFBAgJQjtIwDCWOY1VOMgZpWI/Qdg4gwVhbF7P4aLxhUGz/tqvwC6d+8ilkwi8OAAAAAElFTkSuQmCC"},"images":{"fallback":{"src":"/static/49707739096ba8546170696159e53069/59237/codegrading.png","srcSet":"/static/49707739096ba8546170696159e53069/75651/codegrading.png 175w,\n/static/49707739096ba8546170696159e53069/16036/codegrading.png 350w,\n/static/49707739096ba8546170696159e53069/59237/codegrading.png 700w,\n/static/49707739096ba8546170696159e53069/6f383/codegrading.png 1400w","sizes":"(min-width: 700px) 700px, 100vw"},"sources":[{"srcSet":"/static/49707739096ba8546170696159e53069/92bb7/codegrading.avif 175w,\n/static/49707739096ba8546170696159e53069/a9133/codegrading.avif 350w,\n/static/49707739096ba8546170696159e53069/c70c1/codegrading.avif 700w,\n/static/49707739096ba8546170696159e53069/d9703/codegrading.avif 1400w","type":"image/avif","sizes":"(min-width: 700px) 700px, 100vw"},{"srcSet":"/static/49707739096ba8546170696159e53069/c2861/codegrading.webp 175w,\n/static/49707739096ba8546170696159e53069/64062/codegrading.webp 350w,\n/static/49707739096ba8546170696159e53069/cd1c0/codegrading.webp 700w,\n/static/49707739096ba8546170696159e53069/f4456/codegrading.webp 1400w","type":"image/webp","sizes":"(min-width: 700px) 700px, 100vw"}]},"width":700,"height":462}}},"tech":["Python","Machine Learning","NLP","LLVM"],"github":"https://github.com/orgs/automated-code-analysis-and-grading/dashboard","external":"https://ieeexplore.ieee.org/document/8940623"},"html":"\n- Developed an open source codebase that grades quality and logic of C/C++ codes using feature extraction on a combination of control flow graph and data flow graph. Published a reserach paper documenting our approach.
\n- Evaluated the performance of different ML models (including SVM, Linear Regression, Ridge Regression and Ordinal Regression) on our novel dataset consisting of multiple coding responses to obtain optimal results.
\n
"}},{"node":{"frontmatter":{"title":"Developing GAN for Image-to-Image Translation","cover":{"childImageSharp":{"gatsbyImageData":{"layout":"constrained","placeholder":{"fallback":"data:image/png;base64,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"},"images":{"fallback":{"src":"/static/f8078ca94d65b8e85382d1713b2b2a81/40111/gan.png","srcSet":"/static/f8078ca94d65b8e85382d1713b2b2a81/877f3/gan.png 175w,\n/static/f8078ca94d65b8e85382d1713b2b2a81/17f29/gan.png 350w,\n/static/f8078ca94d65b8e85382d1713b2b2a81/40111/gan.png 700w","sizes":"(min-width: 700px) 700px, 100vw"},"sources":[{"srcSet":"/static/f8078ca94d65b8e85382d1713b2b2a81/4845b/gan.avif 175w,\n/static/f8078ca94d65b8e85382d1713b2b2a81/6a610/gan.avif 350w,\n/static/f8078ca94d65b8e85382d1713b2b2a81/6397d/gan.avif 700w","type":"image/avif","sizes":"(min-width: 700px) 700px, 100vw"},{"srcSet":"/static/f8078ca94d65b8e85382d1713b2b2a81/10109/gan.webp 175w,\n/static/f8078ca94d65b8e85382d1713b2b2a81/b5c5d/gan.webp 350w,\n/static/f8078ca94d65b8e85382d1713b2b2a81/ee4a2/gan.webp 700w","type":"image/webp","sizes":"(min-width: 700px) 700px, 100vw"}]},"width":700,"height":492}}},"tech":["Machine Learning","Tensorflow","Keras","cGANs"],"github":"https://github.com/ansh1204/Image_to_Image_Translation_GAN","external":"https://github.com/ansh1204/Image_to_Image_Translation_GAN/blob/main/Project_Report.pdf"},"html":"\n- Implemented a cGAN architecture trained through an adversarial process for Image-to-Image translation task focusing on satellite to map view task.
\n- Conducted a comparative study between Neural Style Transfer and cGANs to analyse the better performing model for the task in hand.
\n
"}}]}}}