Hi all, I'm Yash 👋

A final year undergrad with a passion for Deep Learning and Machine Learning, actively looking for full-time AI/ML researcher roles. 🚀 I am currently pursuing Mathematics and Computing with a minor in Computer Science from IIT Kharagpur.🚀 I also trade stocks in my free time🚀

saad sitting on table
Saad Working

What i do

A DEEP LEARNING ENTHUSIAST WHO WANTS TO EXPLORE EVERY TECH STACK

  • aws

  • python

  • docker

⚡ Proficient in PyTorch, TensorFlow and Distributed Computing.

⚡ Well versed with the recent development in the feild of Deep Learning, specifically NLP.

⚡ Can deploy ML/AI models on a cloud platform, with a basic frontend.

⚡ Have solid Mathematical and Algorithms Knowledge.

Publications

Papers that I have co-authored

Distributed representation of tags for Active Zero Shot learning

CODS COMAD(submitted)|Extreme multi-labelled classification (XMLC) refers to the problem of tagging items to its most relevant subset of class labels from an extremely large set of labels. Since it is practically difficult to obtain training instances for each of such a large set of labels, the problem is better tackled by a zero-shot approach, where training instances of only a small subset of labels are used. In this paper, we focus on active zero-shot classification, where one can select which subset of labels to obtain training data for. The challenge is to intelligently select a small subset of labels (seen classes) for which to obtain training data, such that the trained model can accurately predict a large number of unseen labels. We present a model named DistributedRepresentation of Tags for Active Zero-Shot Learning (DiRTAZL), which learns label similarities from an external knowledge source, to intelligently select the seen classes. We experimented on three data sets to demonstrate that our proposed model is superior to prior models for active zero-shot learning paradigm

→

MODULATION SPECTRAL SIGNAL REPRESENTATION AND I-VECTORS FOR ANOMALOUS SOUND DETECTION

DCASE 2020| We propose two different anomalous sound detection systems, one based on features extracted from a modulation spectral signal representation and the other based on i-vector extracted from mel-band features. The first system uses a nearest neighbour graph to construct clusters which capture local variations in the training data. Anomalies are then identified based on their distance from the cluster centroids. The second system uses i-vectors extracted from mel-band spectra for training a Gaussian Mixture Model. Anomalies are then identified using their negative log likelihood. Both these methods show significant improvement over the DCASE Challenge baseline AUC scores, with an average improvement of 6% across all machines. An ensemble of the two systems is shown to further improve the average performance by 11% over the baseline.

→

Exploring Causes of Crane Accidents from Incident Reports Using Decision Tree

Electrical Overhead Traveling (EOT) cranes in manufacturing industries serve the purpose of material handling in a complex working environment. The complexity involved in human-machine interaction at the workplace make it hazard and incident prone. In the current study, emerging data mining technique like Decision tree (DT) is adopted to explore the underlying causes involved in incidents that happened in the studied plant from the year 2014–2016. Interesting results are obtained from the analysis like the number of incidents happened during construction and maintenance activities and in weekend (Saturday, Sunday) are more. Managerial implications are suggested for the betterment of the safety management system of the studied plant.

→

Visually Aware Skip-Gram for Image Based Recommendations

WSDM 2021(submitted)|The visual appearance of a product significantly influences purchase decisions on e-commerce websites. We propose a novel framework VASG (Visually Aware Skip-Gram) for learning user and product representations in a common latent space using product image features. Our model is an amalgamation of the Skip-Gram architecture and a deep neural network based Decoder. Here the Skip-Gram attempts to capture user preference by optimizing user-product co-occurrence in a Heterogeneous Information Network while the Decoder simultaneously learns a mapping to transform product image features to the Skip-Gram embedding space. This architecture is jointly optimized in an end-to-end, multitask fashion. The proposed framework enables us to make personalized recommendations for cold-start products which have no purchase history. Experiments conducted on large real world datasets show that the learned embeddings can generate effective recommendations using nearest neighbour searches.

→

Experiences

Goldman Sachs
Goldman Sachs
Risk Analyst
June 2020 – Present

To estimate and validate the intraday liquidity requirement of the firm

    University of Alberta
    University of Alberta
    Deep Learning Research Intern
    May 2019 – July 2019

    • Made a visualization tool to find anomalies and defects in 3D laser printed metal parts using thermal imaging
    • Implemented Deep Predictive Coding Networks and Conv-LSTMs to predict the geometry of the next layer in 3D printed metal samples using thermal imaging, surface profile, 3D point cloud and process parameters
    Yellow Messenger
    Yellow Messenger
    Deep Learning Intern
    May 2018 – Sep 2018

    • Implemented Generative Chatbots using seq2seq modeling in tensorflow and trained it on domain specific data
    • Was responsible for making an Optical Character Recognition model to extract metadata from Id-cards
    • Spearheaded development of a LSTM based model for emotion recognition from human speech

    Projects 🏆

    Projects and Some Cool Stuff that I have done !

    Extreme multi-labeled classification

    • Used Active-Zero Shot Learning for predicting classes using labeled data only from a small set of classes • Used Modified Skip-Gram along with Page Rank as a centrality measure to estimate class relations

    Smart Social Media

    • Implemented LSTMs, convNets and Bag-of-Words using pytorch to classify hate speech on live Twitter feed• The back-end models were hosted on an online sever to use the Chrome extension while scrolling through live Twitter feed. Awarded the Best semester project among 13 groups in a peer review

    Automated Summarization of Legal Text Files

    • Compared different Summarization Algorithms like Text Rank and Lex-Rank for summarizing legal text files • Evaluated the computer made summaries with gold standard summaries using ROUGE metric for scoring

    Vocabulary Building and Query Reformulation for E-commerce

    • Used Conditional Random Field Tagging with LSTMs for sequence tagging of Fashion related articles• Implemented Siamese and Triplet networks using pytorch to tackle the problem of vocabulary mismatching

    Achievements And Certifications 🏆

    Achievements, Certifications, Award Letters

    Inter-IIT Gold Medalist

    Part of 4- member gold wining team, selected by IIT Kharagpur to represent it at TV Audience Measurement event. The presented solution is being currently adopted by Broadcast Audience Research Council, India

    Teaching Assistant Nepal School of AI

    Was selected as a teaching assitant for NLP at the 2nd Nepal School of AI.

    Data Analysis Hackathon, KTJ IIT Kharagpur, Gold Medalist

    Made an app to decide tax fraud using feed from social media

    Blogs

    I love to read and teach others what I have read.

    Contact Me ☎️

    Discuss an idea or just want to say hi? My Inbox is open for all.

    +91-9074521111



    Saad Working

    Made with ❤️ by Yash