index.json

    
    
    
    
    
    
    
    
    
    
    
    
    
    
    
    
    
    
    
    
    
    [{"authors":null,"categories":null,"content":"I am a Machine Learning Engineer at Cruise developing a critical component that predicts trajectories of actors around Cruise’s autonomous vehicle (AV), enabling the car to drive safely, comfortably, and human-like. I am broadly interested at the juncture of deep learning and computer vision, with forays into RL and NLP. I aim to solve challenging and meaningful real-world problems, like self-driving, through the power of AI.\nPreviously, I received my MSE in CIS from the University of Pennsylvania and B. Tech. (Honors) in CS from NIT Trichy. In the past, I have been fortunate to work with amazing mentors, Prof. Kostas Daniilidis (GRASP Lab, Penn), Prof. Eric Granger (LIVIA lab, ETS Montreal), Prof. C. V. Jawahar (CVIT lab, IIIT Hyderabad), and Prof. Venkatesh Babu (VAL Lab, IISc Bangalore).\nI am also a recipient of several prestigious awards like the University of Pennsylvania’s Excellence in Team Leadership, Vector Scholarship in Artificial Intelligence, Mitacs Globalink Graduate Fellowship and Indian Academy of Sciences’ Summer Research Fellowship.\n","date":1684713600,"expirydate":-62135596800,"kind":"term","lang":"en","lastmod":1684713600,"objectID":"2525497d367e79493fd32b198b28f040","permalink":"","publishdate":"0001-01-01T00:00:00Z","relpermalink":"","section":"authors","summary":"I am a Machine Learning Engineer at Cruise developing a critical component that predicts trajectories of actors around Cruise’s autonomous vehicle (AV), enabling the car to drive safely, comfortably, and human-like.","tags":null,"title":"Benedict Florance Arockiaraj","type":"authors"},{"authors":[],"categories":null,"content":" Click on the Slides button above to view the built-in slides feature. Slides can be added in a few ways:\nCreate slides using Wowchemy’s Slides feature and link using slides parameter in the front matter of the talk file Upload an existing slide deck to static/ and link using url_slides parameter in the front matter of the talk file Embed your slides (e.g. Google Slides) or presentation video on this page using shortcodes. Further event details, including page elements such as image galleries, can be added to the body of this page.\n","date":1906549200,"expirydate":-62135596800,"kind":"page","lang":"en","lastmod":1906549200,"objectID":"a8edef490afe42206247b6ac05657af0","permalink":"https://benedictflorance.github.io/talk/example-talk/","publishdate":"2017-01-01T00:00:00Z","relpermalink":"/talk/example-talk/","section":"event","summary":"An example talk using Wowchemy's Markdown slides feature.","tags":[],"title":"Example Talk","type":"event"},{"authors":["Bruce W. Lee","Benedict Florance Arockiaraj","Helen Jin"],"categories":null,"content":"","date":1684713600,"expirydate":-62135596800,"kind":"page","lang":"en","lastmod":1684713600,"objectID":"69425fb10d4db090cfbd46854715582c","permalink":"https://benedictflorance.github.io/publication/conference-paper/","publishdate":"2023-05-22T00:00:00Z","relpermalink":"/publication/conference-paper/","section":"publication","summary":"Truthfulness prediction of LLM's responses using linguistic features.","tags":[],"title":"Linguistic Properties of Truthful Responses","type":"publication"},{"authors":null,"categories":null,"content":" We investigate the phenomenon of an LLM’s untruthful response using a large set of 220 handcrafted linguistic features. We focus on GPT-3 models and find that the linguistic profiles of responses are similar across model sizes. That is, how varying-sized LLMs respond to given prompts stays similar on the linguistic properties level. We expand upon this finding by training support vector machines that rely only upon the stylistic components of model responses to classify the truthfulness of statements. Though the dataset size limits our current findings, we present promising evidence that truthfulness detection is possible without evaluating the content itself. This paper was accepted at TrustNLP @ ACL 2023 ","date":1684108800,"expirydate":-62135596800,"kind":"page","lang":"en","lastmod":1684108800,"objectID":"b913bb4b4575889b031dfb9ed0227d2d","permalink":"https://benedictflorance.github.io/research/acl/","publishdate":"2023-05-15T00:00:00Z","relpermalink":"/research/acl/","section":"research","summary":"Truthfulness prediction of LLM's responses using linguistic features.","tags":["Deep Learning","NLP"],"title":"Linguistic Properties of Truthful Responses","type":"research"},{"authors":null,"categories":null,"content":" Search Engine was built using four key components: crawler, indexer and data processor, ranker and frontend. A corpus of 500k URLs were crawled and efficiently indexed. Sound ranking algorithms were used to rank results with a minimalistic UI to demonstrate the search engine. ","date":1682899200,"expirydate":-62135596800,"kind":"page","lang":"en","lastmod":1682899200,"objectID":"275a899134a8db273e832c2ae6568aaf","permalink":"https://benedictflorance.github.io/project/holi/","publishdate":"2023-05-01T00:00:00Z","relpermalink":"/project/holi/","section":"project","summary":"A Google Search like search engine.","tags":["Software"],"title":"Holi Search Engine","type":"project"},{"authors":null,"categories":null,"content":" Developed a mini-minecraft with various features including procedural generation of terrain, efficient storing and rendering of blocks, player physics, multi-threaded terrain loading, texturing and animation, procedurally-generated caves, day and night cycles, distance fog, dynamic clouds, inventory, water waves, sounds, L-system rivers and NPC AI. ","date":1669852800,"expirydate":-62135596800,"kind":"page","lang":"en","lastmod":1669852800,"objectID":"f7ece7b09a3b78d975c17dc023ade49e","permalink":"https://benedictflorance.github.io/project/computer_graphics/","publishdate":"2022-12-01T00:00:00Z","relpermalink":"/project/computer_graphics/","section":"project","summary":"A Minecraft style interactive 3D world exploration game.","tags":["Graphics"],"title":"Mini Minecraft","type":"project"},{"authors":null,"categories":null,"content":" Presented an excellent fault-tolerant PennCloud platform that supports a complete set of services with an intuitive user interface, including e.g., a webmail service for both local and remote users, a storage service for uploading and downloading of large files in any format, and an admin console for viewing and easy controlling of frontend and backend nodes’ status and data. Foundational to these services, the platform also features a solid, scalable system design with strong consistency, efficient fault detection and recovery, fast performance and great usability. This project was included in the Hall of Fame for being well-done in every aspect. ","date":1669852800,"expirydate":-62135596800,"kind":"page","lang":"en","lastmod":1669852800,"objectID":"788657f7b7270a55087bdf068a81bcc8","permalink":"https://benedictflorance.github.io/project/penn_cloud/","publishdate":"2022-12-01T00:00:00Z","relpermalink":"/project/penn_cloud/","section":"project","summary":"A distributed Google Workspace clone emulating GMail and GDrive.","tags":["Software"],"title":"PennCloud","type":"project"},{"authors":null,"categories":null,"content":" Developed data, pre-processing and modeling pipeline for Transformers, GNNs-based trajectory prediction architecture that acts on vectorized map and agent information. Improved prediction quality through lossless and long-range feature representation, reduced prediction latency by 20%, and improved training and validation speeds by 33%. ","date":1659657600,"expirydate":-62135596800,"kind":"page","lang":"en","lastmod":1659657600,"objectID":"1df1044ce1c3bd301fd9a4dff4abac96","permalink":"https://benedictflorance.github.io/workex/trajectory_prediction/","publishdate":"2022-08-05T00:00:00Z","relpermalink":"/workex/trajectory_prediction/","section":"workex","summary":"Transformers-based trajectory prediction architecture predicting intent of actors around Cruise AV.","tags":null,"title":"Trajectory Prediction for Cruise Autonomous Cars","type":"workex"},{"authors":null,"categories":null,"content":" Designing a model-based reinforcement-learning approach via world models using a novel combination of intrinsic and sparse extrinsic reward for robotic manipulation tasks in MetaWorld and adapting to new tasks exploiting prior experience. Advisors: Kostas Daniilidis and Oleg Rybkin ","date":1652572800,"expirydate":-62135596800,"kind":"page","lang":"en","lastmod":1652572800,"objectID":"a3f2e7cf6474cca31830abb11192014b","permalink":"https://benedictflorance.github.io/research/model_rl/","publishdate":"2022-05-15T00:00:00Z","relpermalink":"/research/model_rl/","section":"research","summary":"Model-based reinforcement-learning approach via world models.","tags":["Deep Learning","RL"],"title":"Unsupervised Reinforcement Learning via World Models","type":"research"},{"authors":null,"categories":null,"content":" Used transfer learning to achieve the fast lap times in OpenAI’s Car racing environment by training the agent on one circuit, and racing it on other customized target environments by zero-shot transfer or by additional fine-tuning. Compared the performance of model-based and model-free approaches, and observed that model-based approaches dominate in performance and converge faster than model-free approaches in this environment. Observed that transfer learning in most setups not only boosts the performance on the target domain, but also shows high performance ability during learning. ","date":1650758400,"expirydate":-62135596800,"kind":"page","lang":"en","lastmod":1650758400,"objectID":"d2ef76bb43fbdf5b54c982859de1277b","permalink":"https://benedictflorance.github.io/project/transfer_rl/","publishdate":"2022-04-24T00:00:00Z","relpermalink":"/project/transfer_rl/","section":"project","summary":"Model-free \u0026 model-based transfer learning for car-racing.","tags":["ML"],"title":"Transfer Learning for Car-Racing Environments","type":"project"},{"authors":null,"categories":null,"content":" Extended density-map estimation based FamNet to count highly-occluded machine parts with a novel mismatch loss component, achieving a performance of 1.96 MAE (90% decrease from image processing baseline) on the challenge dataset. ","date":1640563200,"expirydate":-62135596800,"kind":"page","lang":"en","lastmod":1640563200,"objectID":"51c778b9413297f0f44020445562c876","permalink":"https://benedictflorance.github.io/project/counting_parts/","publishdate":"2021-12-27T00:00:00Z","relpermalink":"/project/counting_parts/","section":"project","summary":"Industrial challenge to count highly occluded washer parts.","tags":["ML"],"title":"Counting Machine Washer Parts","type":"project"},{"authors":null,"categories":null,"content":" Proposed a novel technique combining parts-of-speech filtering and perplexity based loss function to generate sensible triggers that are closer to natural phrases. For the task of sentiment analysis on the SST dataset, the method produced sensible triggers that achieve accura- cies as low as 4% and 12% for flipping positive to negative predictions and vice-versa. To build robust models, performed adversarial training using the generated triggers that increases the accuracy of the model from 12% to 48%. Illustrated that adversarial at- tacks can be made difficult to detect by generating sensible triggers, and to facilitate robust model development through relevant defenses. ","date":1640563200,"expirydate":-62135596800,"kind":"page","lang":"en","lastmod":1640563200,"objectID":"f6412dd43ab403cb54eaab31688296a5","permalink":"https://benedictflorance.github.io/research/adversarial_triggers/","publishdate":"2021-12-27T00:00:00Z","relpermalink":"/research/adversarial_triggers/","section":"research","summary":"Generating sensible universal adversarial triggers that are closer to natural phrases.","tags":["Deep Learning","NLP"],"title":"Sensible Universal Adversarial Triggers","type":"research"},{"authors":null,"categories":null,"content":" Built deep-learning pipelines using object detection, segmentation, fine-grained classification and self-supervised learning for retailers like Kimberly Clark, P\u0026amp;G, Lowe’s, Coke and ABInBev to provide real-time competitive intelligence and on-shelf execution insights. Achieved \u0026gt;97% accuracy in detecting the smallest of SKUs and lifted per-store sales by 5%. ","date":1625961600,"expirydate":-62135596800,"kind":"page","lang":"en","lastmod":1625961600,"objectID":"3a00e16f1867fe68d248443b86e54cd2","permalink":"https://benedictflorance.github.io/workex/infilect/","publishdate":"2021-07-11T00:00:00Z","relpermalink":"/workex/infilect/","section":"workex","summary":"Advanced image recognition and retail execution analytics for worldwide retail brands.","tags":null,"title":"Visual Intelligence Analytics for FMCG Brands","type":"workex"},{"authors":null,"categories":null,"content":" Wrote data-loaders and modeled the architecture for kinematic-structure preserving, unsupervised 3D pose estimation framework to effectively disentangle pose, foreground and background appearance information.\nReduced MPJPE by as high as 40% (semi-supervised) and 15% (unsupervised) on datasets like Human3.6M, 3DHP, LSP and 3DPW.\nAdvisors: Venkatesh Babu and Jogendra Nath Kundu\n","date":1598745600,"expirydate":-62135596800,"kind":"page","lang":"en","lastmod":1598745600,"objectID":"271b2502e0aa7acf8dbbb520ba8f8287","permalink":"https://benedictflorance.github.io/research/iisc/","publishdate":"2020-08-30T00:00:00Z","relpermalink":"/research/iisc/","section":"research","summary":"A kinematic-structure-preserving unsupervised 3D pose estimation framework.","tags":["Deep Learning","Computer Vision"],"title":"Unsupervised 3D Human Pose Estimation","type":"research"},{"authors":null,"categories":null,"content":" Proposed an ensemble of RetinaNet-based object detectors to localize bounding boxes and predict labels of eight different artefact classes that generalizes to an inter-patient, multi-tissue and a multi-modal corpus of endoscopy video frame data\nThe common artefacts of interest that corrupt endoscopy video frames include contrast, saturation, instrument, blood, specularity, blur, imaging artefacts and bubbles.\nAchieved an mAP of 0.3405 improving existing state-of-the-art results\nAdvisor: Leela Velusamy\n","date":1590796800,"expirydate":-62135596800,"kind":"page","lang":"en","lastmod":1590796800,"objectID":"471537725756ddd81296bc0aab78824c","permalink":"https://benedictflorance.github.io/project/nitt/","publishdate":"2020-05-30T00:00:00Z","relpermalink":"/project/nitt/","section":"project","summary":"Ensemble-based endoscopy video artefact detection.","tags":["ML"],"title":"Ensemble-based Endoscopy Artefact Detection","type":"project"},{"authors":null,"categories":null,"content":" Analyzed negative transfer (around 20% drop in mAP from baseline) and catastrophic forgetting of the existing imageto-image domain adaptation approaches on face-detection datasets. Studied the use of local features, and temporal information from trackers to generalize unsupervised domain adaptation approaches on datasets like SCUT and Widerface. Advisor: Eric Granger ","date":1565827200,"expirydate":-62135596800,"kind":"page","lang":"en","lastmod":1565827200,"objectID":"6fb6b93d778dc6f5a8031089943b336f","permalink":"https://benedictflorance.github.io/research/uquebec/","publishdate":"2019-08-15T00:00:00Z","relpermalink":"/research/uquebec/","section":"research","summary":"Analysis of the current image-to-image adpation methods on video-surveillance datasets.","tags":["Deep Learning","Computer Vision"],"title":"Unsupervised Deep Domain Adaption for Object Detection","type":"research"},{"authors":null,"categories":null,"content":" Employed video frame segmentation, replay and scoreboard detection on a newly constructed dataset of IPL cricket videos.\nConducted a study on the real-world implementation of sports analytics by interviewing Mr. K.S. Viswanathan, CEO of Chennai Super Kings, the most successful franchise cricket team and other experts from the domain\n","date":1559174400,"expirydate":-62135596800,"kind":"page","lang":"en","lastmod":1559174400,"objectID":"e74b17dda6ecf3f4c51e2c3ef59534c8","permalink":"https://benedictflorance.github.io/project/cricket/","publishdate":"2019-05-30T00:00:00Z","relpermalink":"/project/cricket/","section":"project","summary":"Automatic highlights generation for broadcast cricket videos.","tags":["ML"],"title":"Highlights Generation for Cricket Videos","type":"project"},{"authors":[],"categories":[],"content":"Create slides in Markdown with Wowchemy Wowchemy | Documentation\nFeatures Efficiently write slides in Markdown 3-in-1: Create, Present, and Publish your slides Supports speaker notes Mobile friendly slides Controls Next: Right Arrow or Space Previous: Left Arrow Start: Home Finish: End Overview: Esc Speaker notes: S Fullscreen: F Zoom: Alt + Click PDF Export Code Highlighting Inline code: variable\nCode block:\nporridge = \u0026#34;blueberry\u0026#34; if porridge == \u0026#34;blueberry\u0026#34;: print(\u0026#34;Eating...\u0026#34;) Math In-line math: $x + y = z$\nBlock math:\n$$ f\\left( x \\right) = ;\\frac{{2\\left( {x + 4} \\right)\\left( {x - 4} \\right)}}{{\\left( {x + 4} \\right)\\left( {x + 1} \\right)}} $$\nFragments Make content appear incrementally\n{{% fragment %}} One {{% /fragment %}} {{% fragment %}} **Two** {{% /fragment %}} {{% fragment %}} Three {{% /fragment %}} Press Space to play!\nOne Two Three A fragment can accept two optional parameters:\nclass: use a custom style (requires definition in custom CSS) weight: sets the order in which a fragment appears Speaker Notes Add speaker notes to your presentation\n{{% speaker_note %}} - Only the speaker can read these notes - Press `S` key to view {{% /speaker_note %}} Press the S key to view the speaker notes!\nOnly the speaker can read these notes Press S key to view Themes black: Black background, white text, blue links (default) white: White background, black text, blue links league: Gray background, white text, blue links beige: Beige background, dark text, brown links sky: Blue background, thin dark text, blue links night: Black background, thick white text, orange links serif: Cappuccino background, gray text, brown links simple: White background, black text, blue links solarized: Cream-colored background, dark green text, blue links Custom Slide Customize the slide style and background\n{{\u0026lt; slide background-image=\u0026#34;/media/boards.jpg\u0026#34; \u0026gt;}} {{\u0026lt; slide background-color=\u0026#34;#0000FF\u0026#34; \u0026gt;}} {{\u0026lt; slide class=\u0026#34;my-style\u0026#34; \u0026gt;}} Custom CSS Example Let’s make headers navy colored.\nCreate assets/css/reveal_custom.css with:\n.reveal section h1, .reveal section h2, .reveal section h3 { color: navy; } Questions? Ask\nDocumentation\n","date":1549324800,"expirydate":-62135596800,"kind":"page","lang":"en","lastmod":1549324800,"objectID":"0e6de1a61aa83269ff13324f3167c1a9","permalink":"https://benedictflorance.github.io/slides/example/","publishdate":"2019-02-05T00:00:00Z","relpermalink":"/slides/example/","section":"slides","summary":"An introduction to using Wowchemy's Slides feature.","tags":[],"title":"Slides","type":"slides"},{"authors":null,"categories":null,"content":" Performed transfer learning on various CNN models using our new Indic-Leaf dataset of 112 Indian plants for identification and disease detection with results showing Top-1 and Top-5 precision of 90.08 and 96.90\nDeployed auto-segmenter, auto-recommender systems and a collaborative web-portal using Angular and NodeJS and an Android app where users can add new observations and query for species identification.\nAdvisor: C. V. Jawahar\n","date":1532908800,"expirydate":-62135596800,"kind":"page","lang":"en","lastmod":1532908800,"objectID":"2b480d3f86125ce25026d240a59a5df9","permalink":"https://benedictflorance.github.io/project/iiith/","publishdate":"2018-07-30T00:00:00Z","relpermalink":"/project/iiith/","section":"project","summary":"Plant identification and disease detection system.","tags":["ML"],"title":"Indian Flora Project","type":"project"},{"authors":null,"categories":null,"content":" Developed the three-day online mathematical challenge MDecoder, where contestants answer mathematical questions of different difficulty levels from various disciplines of mathematics like algebra, geometry to applied maths Built the web application using Angular library for the user interface, PHP framework Laravel for the backend service and hosted in Docker-based environment Implemented efficient, foolproof algorithms for authentication, time-based scoring, question generation and submission validation The contest was held as a part of Pragyan, the technical festival of NIT Trichy and it witnessed over 1000+ contestants and 15,000+ submissions. ","date":1525478400,"expirydate":-62135596800,"kind":"page","lang":"en","lastmod":1525478400,"objectID":"b47b00e9afe49e2cad3fee31278ae3fc","permalink":"https://benedictflorance.github.io/project/mdecoder/","publishdate":"2018-05-05T00:00:00Z","relpermalink":"/project/mdecoder/","section":"project","summary":"Online math contest in Pragyan, tech-fest of NITT.","tags":["Software"],"title":"M-Decoder (Online Math Challenge)","type":"project"},{"authors":null,"categories":null,"content":" Developed the backend service for digitizing the records of inventory management, production planning and status updates of the coaches and rakes under the Digital India initiative Wrote API’s in Laravel framework to handle authorization across Web and Android clients The code was reviewed and pushed to production. The Android application has been released on Google Play Store and is used by more than 500 employees of ICF, Chennai for internal purposes. ","date":1515110400,"expirydate":-62135596800,"kind":"page","lang":"en","lastmod":1515110400,"objectID":"b076f133ddd2876cf6142418f8512209","permalink":"https://benedictflorance.github.io/project/icf/","publishdate":"2018-01-05T00:00:00Z","relpermalink":"/project/icf/","section":"project","summary":"Coach and Rake production tracker for Indian Railways.","tags":["Software"],"title":"Indian Railways EMU Production App","type":"project"},{"authors":null,"categories":null,"content":" Implemented deep learning models to predict disease risks based on user symptoms and lifestyle inputs as a collaborative pipeline between doctors and patients. Used web-scraped data for training the model Built a recommendation system for analyzing the abnormal parameters and recommending future course of action like medicines and wrote APIs in Laravel for the Android application where the features were deployed This project received a Special Mention at the Pragyan Hackathon conducted at SAP Labs, Bangalore. ","date":1483574400,"expirydate":-62135596800,"kind":"page","lang":"en","lastmod":1483574400,"objectID":"dd259528993d9e0aa75010d02122d514","permalink":"https://benedictflorance.github.io/project/healthcare/","publishdate":"2017-01-05T00:00:00Z","relpermalink":"/project/healthcare/","section":"project","summary":"A smart healthcare system for doctors and patients.","tags":["Software"],"title":"Smart Health Care System","type":"project"}]