/ Recent content in Home on Tamanna Hossain-Kay Hugo -- gohugo.io en-us Tue, 27 Jan 2026 00:00:00 +0000 /post/2026/01/27/satlm/ Tue, 27 Jan 2026 00:00:00 +0000 /post/2026/01/27/satlm/ Paper:SATLM: Satisfiability-Aided Language Models Using Declarative Prompting Authors: Xi Ye, Qiaochu Chen, Isil Dillig, Greg Durrett Chain-of-thought works best when the task is mostly forward (has a linear procedure), and adding tools (e.g., a Python executor) helps most when the main failure mode is arithmetic or other execution mistakes. But many real problems are closer to constraint solving: you need to pick a plan, search over possibilities, and keep global consistency—where LLMs often fail due to planning errors even if they can execute steps correctly. /post/2025/02/08/deepseek/ Sat, 08 Feb 2025 00:00:00 +0000 /post/2025/02/08/deepseek/ The release of DeepSeek-R1, a Chinese open-weight language model, in January 2025 sent shockwaves through the tech industry, triggering a stunning $1 trillion loss in US stock market value in a single day. For the first time in a few years, an open-weight model has demonstrated the capability to rival proprietary giants like OpenAI’s GPT models. Not only does it offer competitive performance, but it also achieves this using older GPUs than Western labs due to US export restrictions. /post/2023/08/05/paper-summary-whose-opinions-do-language-models-reflect/ Sat, 05 Aug 2023 00:00:00 +0000 /post/2023/08/05/paper-summary-whose-opinions-do-language-models-reflect/ Paper Link: https://arxiv.org/pdf/2303.17548.pdf Authors: Shibani Santurkar \(^1\), Esin Durmus\(^1\), Faisal Ladhak\(^2\), Cinoo Lee\(^1\), Percy Liang\(^1\), Tatsunori Hashimoto\(^1\) (\(^1\) Stanford, \(^2\) Columbia University) Language models, or LMs, can give “opinionated” answers to open-ended, subjective questions. But whose opinions are these? This is important to understand as LMs become more integrated into open-ended applications. Recent studies have shown that LMs can exhibit specific political stances and even mirror beliefs of certain demographics. /post/2022/01/08/paper-summary-harms-of-gender-exclusivity-and-challenges-in-non-binary-representation-in-language-technologies/ Sat, 08 Jan 2022 00:00:00 +0000 /post/2022/01/08/paper-summary-harms-of-gender-exclusivity-and-challenges-in-non-binary-representation-in-language-technologies/ Paper link: https://aclanthology.org/2021.emnlp-main.150.pdf Authors: Sunipa Dev (she/her), Masoud Monajatipoor (he/him), Anaelia Ovalle (they/he/she), Arjun Subramonian (they/them), Jeff M Phillips (he/him), Kai-Wei Chang (he/him) (UCLA) Content Warning: This paper contains examples of stereotypes and associations, misgendering, erasure, and other harms that could be offensive and triggering to trans and nonbinary individuals. Gender in this paper refers to gender identity, which concerns how individuals experience their own gender. Gender identity is broader than the gender binary that is historically normative in Western cultures, i. /post/2021/11/28/weird-stuff-in-high-dimensions/ Sun, 28 Nov 2021 00:00:00 +0000 /post/2021/11/28/weird-stuff-in-high-dimensions/ -- figure { padding-top: 20px; padding-bottom: 20px; } We’re usually putzing around in a \(3\) dimensional space, perceiving the \(4^{th}\) dimension of time as something we’re moving forward through. Attempts to explain our reality, however, have intriguingly and disturbingly given rise to the possibility of higher dimensions. Superstring Theory, for example, posits \(10\) dimensions to provide a unifying framework between general relativity and quantum mechanics. M-Theory, on the other hand, posits \(11\) dimensions, and Bosonic String Theory a daunting \(26\) dimensions! /post/2021/11/09/box-embeddings-paper-overviews/ Tue, 09 Nov 2021 00:00:00 +0000 /post/2021/11/09/box-embeddings-paper-overviews/ Geometric embeddings can be used to represent transitive asymmetric relationships. Box-embeddings are n-dimensional hyperrectangles that can represent such relationships by containment. Source: Box Embeddings: An open-source library for representation learning using geometric structures Formally, a “box” is defined as a Cartesian product of closed intervals, \[ \begin{aligned} B(\theta) &=\prod_{i=1}^{n}\left[z_{i}(\theta), Z_{i}(\theta)\right] \\ &=\left[z_{1}(\theta), Z_{1}(\theta)\right] \times \cdots \times\left[z_{n}(\theta), Z_{n}(\theta)\right]\\ &=[x_{m,1},x_{M,1}]\times \cdots \times [x_{m,n},x_{M,n}] \end{aligned} \] Parameterization: \(\theta\) may be the output from a neural network /post/2021/06/28/orthogonal-procrustes/ Mon, 28 Jun 2021 00:00:00 +0000 /post/2021/06/28/orthogonal-procrustes/ I came across the Orthogonal Procrustes technique when reading the paper Diachronic Word Embeddings Reveal Statistical Laws of Semantic Change. It was used to align word embeddings from different runs of SVD (Singular Value Decomposition) and SGNS (Skipgram with Negative Sampling), since they can both result in arbitrary orthogonal transformations of word vectors. So what is Orthogonal Procrustes? Problem Solution Proof of Solution Example Problem Given two matrices \(A\), \(B\) find a matrix \(R\) for mapping \(A\) to \(B\). /post/2019/12/25/five-year-anniversary-trip-to-maui/ Wed, 25 Dec 2019 00:00:00 +0000 /post/2019/12/25/five-year-anniversary-trip-to-maui/ My husband and I got married on Christmas Eve, 2014, and this year we decided to celebrate our fifth anniversary vacationing on the island of Maui in Hawaii. Although we live in Irvine, we flew out of San Jose In Northern California because (i) flight prices were cheaper and (ii) upon our return we could easily drive to Santa Cruz to spend some time with my in-laws. /about/ Sat, 28 Sep 2019 00:00:00 +0000 /about/ Some things I like: Science fiction Colors (currently obsessed with pastels) Intersectional feminism Nature stuff (hiking, camping, gardening, beach lounging) Hanging out at breweries Doodling, drawing + consuming visual art Napping Contact Information Email : tamanna.hossainkay@gmail.com LinkedIn : tamanna-hossain-kay Github : atleastbeth3 Instagram : tamannahossainkay Twitter : thossainkay Tableau : tamanna.hossain.kay /post/2018/11/01/a-brief-introduction-to-fmri-analysos/ Thu, 01 Nov 2018 00:00:00 +0000 /post/2018/11/01/a-brief-introduction-to-fmri-analysos/ There are quite a few popular neural imaging modalities in use today. However, they vary in what they measure, level of invasiveness, and spatiotemporal resolution. Functional magnetic resonance imaging has become the dominant technique in recent years because it is noninvasive while also providing decent spatial and temporal resolutions. During an fMRI experiment, brain images are acquired over time while the subject is exposed to certain stimuli or performs a set of tasks. /post/2018/05/18/viz-the-rohingya-exodus/ Fri, 18 May 2018 00:00:00 +0000 /post/2018/05/18/viz-the-rohingya-exodus/ August 25th, 2017 marked a new wave of violence against the Rohingya in Myanmar . Since then over 640,000 refugees have fled to neighboring Bangladesh. The legal condition of the Rohingya has been compared to apartheid by many, including Amnesty International and Nobel Laureate, Desmund Tutu. The United Nations has described them as one of the most persecuted minorities in the world and a UN backed court in Italy fund Myanmar guilty of genocide in September 2017. /post/2018/05/16/viz-drawing-e/ Wed, 16 May 2018 00:00:00 +0000 /post/2018/05/16/viz-drawing-e/ The mathematical constant e is visualized in five different ways below: Parametric (dynamic) uses a parametric equation from Frank Farris’ book Creating Symmetry. A dynamic graph can be found on Sine of the Times. Radial with Transition Lines, Parallel with Transition Lines, Spiral inpired by Ken Flerlage’s Beauty of Pi viz. I needed a geometry refresher to plot points equidistant on an Archimedean Spiral. Radial Rays chart just because I’ve been wanting to try one out in Tableau. /post/2018/02/22/viz-us-same-sex-marriage-laws/ Thu, 22 Feb 2018 00:00:00 +0000 /post/2018/02/22/viz-us-same-sex-marriage-laws/ var divElement = document.getElementById('viz1519290060404'); var vizElement = divElement.getElementsByTagName('object')[0]; vizElement.style.width='1000px';vizElement.style.height='1027px'; var scriptElement = document.createElement('script'); scriptElement.src = 'https://public.tableau.com/javascripts/api/viz_v1.js'; vizElement.parentNode.insertBefore(scriptElement, vizElement); /post/2018/02/11/the-words-of-larry-nassar-survivors/ Sun, 11 Feb 2018 00:00:00 +0000 /post/2018/02/11/the-words-of-larry-nassar-survivors/ More than two hundred women and girls submitted or read impact statements in court regarding sexual abuse they faced at the hand of former USA gymnastics and Michigan State University doctor, Larry Nassar. Their testimonies contributed to his prison sentence on ten counts of sexual assault.These are their words. Testimonies were scraped from the Glamour article ‘My Voice Matters’ : The Survivors of Larry Nassar, In Their Own Words . /post/2017/12/22/encoded-photo-an-anniversary-card/ Fri, 22 Dec 2017 00:00:00 +0000 /post/2017/12/22/encoded-photo-an-anniversary-card/ var divElement = document.getElementById('viz1513973903457'); var vizElement = divElement.getElementsByTagName('object')[0]; vizElement.style.width='800px';vizElement.style.height='777px'; var scriptElement = document.createElement('script'); scriptElement.src = 'https://public.tableau.com/javascripts/api/viz_v1.js'; vizElement.parentNode.insertBefore(scriptElement, vizElement); For my husband on our third anniversary. Encoded in binary in this photo of us is three love poems that I really like - one by Yeats, two by Robert Graves. All required code is on Github. The density of the image matrix is varied along a frame boundary allowing the bits constructing the image to peak through. /post/2017/12/03/text-mining-stranger-things-season-1-vs-season-2/ Sun, 03 Dec 2017 00:00:00 +0000 /post/2017/12/03/text-mining-stranger-things-season-1-vs-season-2/ Right after my husband and I finished binge watching Season 2 , a friend sent me a really interesting sentiment analysis of Stranger Things by Jordan Dworkin. I loved it so much I did a remix ! var divElement = document.getElementById('viz1512538944513'); var vizElement = divElement.getElementsByTagName('object')[0]; vizElement.style.width='800px';vizElement.style.height='1277px'; var scriptElement = document.createElement('script'); scriptElement.src = 'https://public.tableau.com/javascripts/api/viz_v1.js'; vizElement.parentNode.insertBefore(scriptElement, vizElement); The viz is set up as a comparison of episodes across seasons . /post/2017/11/19/metoo-twitter-viz/ Sun, 19 Nov 2017 00:00:00 +0000 /post/2017/11/19/metoo-twitter-viz/ #MeToo is a social media campaign to raise awareness about the prevalence of sexual harassment and violence. Originally coined by the founder of Just Be Inc., Tarana Burke, the slogan went viral in October, 2017 after being tweeted by actress Alyssa Milano. As someone who’s both heard and said “me too” many times , I was curious about the global conversation that followed. The following visualization uses a sample of 7000 tweets from October 17, 2017. /post/2017/11/19/metoo-twitter-viz/ Sun, 19 Nov 2017 00:00:00 +0000 /post/2017/11/19/metoo-twitter-viz/ #MeToo is a social media campaign to raise awareness about the prevalence of sexual harassment and violence. Originally coined by the founder of Just Be Inc., Tarana Burke, the slogan went viral in October, 2017 after being tweeted by actress Alyssa Milano. As someone who’s both heard and said “me too” many times , I was curious about the global conversation that followed. The following visualization uses a sample of 7000 tweets from October 17, 2017. /ai_art/ Mon, 01 Jan 0001 00:00:00 +0000 /ai_art/ CLIP Paper: Learning Transferable Visual Models From Natural Language Supervision CLIP for Text-to-Image Generation Gif Source BigSleep Implementaiton Source: BigSleep Repo Portfolio: BigSleep Twitter BigSleep is a combinmation of BigGan and CLIP created by Ryan Murdock, a graduate student studying cognitive neuroscience at the University of Utah.The Big Sleep’s optimization objective when generating images is to find a point in GAN latent space that maximally corresponds to a given sequence of words under CLIP. /dataviz/ Mon, 01 Jan 0001 00:00:00 +0000 /dataviz/ Posts • 2018/11/01 A Brief Introduction to fMRI Analysis 2018/05/18 Viz : The Rohingya Exodus 2018/05/16 Viz : Drawing [e] 2018/02/22 Viz : US Same Sex Marriage Laws 2018/02/11 Viz : The Words of Larry Nassar Survivors 2017/12/22 Viz : Encoded Photo - An Anniversary Card 2017/12/03 Viz : Text Mining Stranger Things (Season 1 vs. /travel/ Mon, 01 Jan 0001 00:00:00 +0000 /travel/ Posts • 2019/12/25 Five Year Anniversary Trip to Maui