Document summarization has greatly benefited from advances in large language models (LLMs). In real-world situations, summaries often need to be generated from multiple documents with diverse sources and authors, lacking a clear information flow. Naively concatenating these documents and generating a summary can lead to poorly structured narratives and redundancy. Additionally, attributing each part of the generated summary to a specific source is crucial for reliability. In this study, we address multi-document summarization with attribution using our proposed solution ***MiDAS-PRo***, consisting of three stages: (i) Planning the hierarchical organization of source documents, (ii) Reasoning by generating relevant entities/topics, and (iii) Summary Generation. We treat the first two sub-problems as a code completion task for LLMs. By incorporating well-selected in-context learning examples through a graph attention network, LLMs effectively generate plans and reason topics for a document collection. Experiments on summarizing scientific articles from public datasets show that our approach outperforms state-of-the-art baselines in both automated and human evaluations.

Abhilash Nandy, Yash Agarwal, Ashish Patwa, Millon Madhur Das, Aman Bansal, Ankit Raj, Pawan Goyal, Niloy Ganguly. Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing. 2024.

A law practitioner has to go through numerous lengthy legal case proceedings for their practices of various categories, such as land dispute, corruption, etc. Hence, it is important to summarize these documents, and ensure that summaries contain phrases with intent matching the category of the case. To the best of our knowledge, there is no evaluation metric that evaluates a summary based on its intent. We propose an automated intent-based summarization metric, which shows a better agreement with human evaluation as compared to other automated metrics like BLEU, ROUGE-L etc. in terms of human satisfaction. We also curate a dataset by annotating intent phrases in legal documents, and show a proof of concept as to how this system can be automated.

Allergic Reactions can range from mild rashes to severe conditions, sometimes even leading to anaphylaxis and sudden death. Lack of sufficient prior patient data and a need of physician's immediate supervision calls for a solution which can be deployed at large to prevent the sudden death occurring due to anaphylactic shocks and further collect data to enable fast detection in future situations. This paper describes an integrated ecosystem comprising of several on-body patient devices connected to a central server with a doctor at one of the client nodes. An on-body device consists of physiological signal acquiring sensors, abnormality detector, and smart-phone for uploading the anomalous data to a server for further classification. Gathering anomalous data from the patients, the cloud processes them through a binary adversarial classifier based on physician's annotation of anaphylaxis occurrence. The adversarial classifier has been incorporated to tackle data insufficiency because of its faster convergence.

Recognizing Families In the Wild (RFIW) is a large-scale kinship recognition challenge based on the FIW dataset. This dataset is the largest databases for kinship recognition, consisting of more than 13,000 family photos and 1,000 families. The number of members in each family range from 4 to 38. One of the tasks for the database is, given photos of two individuals, predict whether they have any kin relationship or not. In this paper, we present a deep learning approach using Siamese Convolutional Neural Network Architecture to quantify the similarity between two given photos. We use two parallel SqueezeNet Networks, initialized with weights obtained after training the SqueezeNet on the VGGFace2 Dataset, and use a similarity metric and fully connected networks to merge the two networks to a single output. We use different similarity metric such as L1 norm, L2 Norm, and Cosine Similarity. Our network gives good accuracy and AUC scores.