I am a member of Microsoft Research lab (AI) in Redmond, Washington.
My research interests are in the areas of Deep Learning, Natural Language Processing and Grounding (Language+Vision). Currently I am excited about following directions, if you are also interested and looking for research collaboration please message me:
- Understanding Errors, Biases and Out Of Distribution (OOD) behavior of Deep Learning models performing Language Understanding and Generation. For example 1 and section V of 2.
- Grounding Language Understanding and Generation tasks in our Visual Experiences, for example, Text Generation (Image/Video Captioning), Retrieval (Image-Text Retrieval), Reasoning (Visual Question Answering).
I have experience working on Linear Inverse Problems [with a focus on Sparse Decomposition and Compressive Sensing], Deep Learning methods for Speech Recognition (2013) and Sentence Modeling for Web Search Engines and Information Retrieval (2014, IEEE Signal Processing Society Best Paper award 2018). I also have worked as a mentor at Microsoft AI School advanced projects class (AI-611) and Microsoft AI Residency Program.
If you are a Ph.D. student and interested in working with me as a research intern please drop me an email at firstname.lastname@example.org.
-  Elevated to IEEE Senior Member.
-  Gave a talk at Facebook AI about “Are Neuro-Symbolic Representations Helpful?“.
-  Will be serving as Associate Editor for IEEE Signal Processing Magazine Newsletter.
-  Are the tasks/datasets that machine learning researchers use for natural language reasoning grounded in vision (visual question answering) actually measure the `reasoning’ capability of the models? Or they are more of a competition about who has a better vision backbone (`perception’) that might be publicly available for others or not? Our recent work proposing a Neuro-Symbolic approach that disentangles reasoning from perception to address this issue has been accepted at ICML 2020.
-  My interview with IEEE Signal Processing Newsletter.
-  Our TP-N2F work is accepted at ICML 2020.
-  Congrats to our intern and all authors of our work (TP-N2F) for the Best Paper Award at NeurIPS 2019’s KR2ML Workshop. MSFT Blog from Paul Smolensky introducing two related projects including this work.
-  Capturing the shared structure among different NLP datasets is the key to achieve meaningful transferability among them. We show that leveraging Neuro-Symbolic representations to disentangle data-specific semantics from general language structure is the key to achieve this, where our proposed model, HUBERT, but not BERT, is able to learn and leverage more effectively. Check it out here.
-  Will be serving as Area Chair for Multimodality track at ACL 2020.
-  How can we leverage the large number of image-text pairs available on the web to mimic the way people improve their scene and language understanding through weak supervision? Our work on large scale Vision and Language Pretraining (VLP) is a step towards this direction. You can download the codes, give it a try! Two short posts related to this work are available at MSFT Blog and VentureBeat. We will be presenting this work at AAAI 2020 (spotlight).
-  Leveraging Neuro-Symbolic representations to solve math problems helps us to better understand neural models and impose necessary discrete inductive biases into them. What are the necessary ingredients for these types of structures for being efficient in these reasoning tasks? Our recent work (TP-N2F) proposes one! Part of our initial results for this work will be presented at NeurIPS 2019’s KR2ML workshop.
-  How large scale Neural Scene Graph Parsers that create a Neuro-Symbolic representation for images can benefit challenging downstream tasks like Text-Image Retrieval and Image Captioning? Check out our effort to address this question.
-  Will be serving as a Member of Organizing Committee of ACL 2020.
-  Our journal paper (published at IEEE/ACM TASLP) on sentence embedding for web search and IR that we did in summer 2014 was selected for IEEE Signal Processing Society 2018 Best Paper award (Test of Time). It was announced during ICASSP 2019 in Brighton, UK. Congratulations to the team and wonderful collaborators!
-  You can read my ICASSP 2019 recap here.
-  Epilepsy is one of the most common neurological disorders in the world, affecting over 70 million people globally, 30 to 40 per cent of whom do not respond to medication. Our recent work published at Clinical Neurophysiology Journal proposes optimized DNN architectures to detect them.
-  Our work on perceptually de-hashing image hashes for similarity retrieval will appear at Signal Processing Image Communications Journal.
-  Our work on robust detection of epileptic seizures will be presented at ICASSP 2018.
-  Our work on leveraging Neuro-Symbolic representations to design DL models with higher interpretability capabilities using Paul Smolensky’s Tensor Product Representations (TPRs) got accepted at AAAI 2018 (Oral Presentation). You can download the paper here. Two short posts related to this work are available here and here.
-  Presented a tutorial at IEEE GlobalSIP 2017 @ Montreal about DL tools and frameworks. If you are similar to me, having the question “which DL framework should I choose for my project?,” check it out here.
-  Our recent results leveraging Neuro-Symbolic representations in deep NLP models will be presented at NIPS 2017’s explainable AI workshop.
-  I have summarized what I learned from Deep Learning Summer School 2016 at Montreal, check it out here.
-  Our recent work on sentence embedding for web search and IR will appear in IEEE/ACM Transactions on Audio, Speech, and Language Processing.
-  Our recent work proposing a deep learning approach for distributed compressive sensing will appear in IEEE Transactions on Signal Processing. We proposed to treat location of each non-zero component as a Symbol and using a Neural representation to find these symbols. We then used least squares to find the non-zero values. Check out the paper and a post about it at Nuit Blanche. The codes are open source, give it a try! For more information about compressive sensing check out here.