Built site for gh-pages

.nojekyll

@@ -1 +1 @@
-7f19e7c7
+b1740025

homeworks.html

@@ -283,9 +283,7 @@
     <h2 id="toc-title">Table of contents</h2>
 
   <ul>
-  <li><a href="#sec-hw-major" id="toc-sec-hw-major" class="nav-link active" data-scroll-target="#sec-hw-major">Major Project</a></li>
-  <li><a href="#sec-hw-minor" id="toc-sec-hw-minor" class="nav-link" data-scroll-target="#sec-hw-minor">Minor Project</a></li>
-  <li><a href="#sec-hw-06" id="toc-sec-hw-06" class="nav-link" data-scroll-target="#sec-hw-06">HW-06</a></li>
+  <li><a href="#sec-hw-06" id="toc-sec-hw-06" class="nav-link active" data-scroll-target="#sec-hw-06">HW-06</a></li>
   <li><a href="#sec-hw-05" id="toc-sec-hw-05" class="nav-link" data-scroll-target="#sec-hw-05">HW-05</a></li>
   <li><a href="#sec-hw-04" id="toc-sec-hw-04" class="nav-link" data-scroll-target="#sec-hw-04">HW-04</a></li>
   <li><a href="#sec-hw-03" id="toc-sec-hw-03" class="nav-link" data-scroll-target="#sec-hw-03">HW-03</a></li>
@@ -316,62 +314,6 @@ <h1 class="title">Homeworks</h1>
 </header>
 
 
-<section id="sec-hw-major" class="level2">
-<h2 class="anchored" data-anchor-id="sec-hw-major">Major Project</h2>
-<p><strong>Task</strong></p>
-<ul>
-<li>This major project focuses on baking all the software engineering best practices, MLOps lifecycle and holistic ML into one code base. All the objectives of the <a href="projects.html#sec-hw-minor" class="quarto-xref"><span>Minor Project</span></a> should be there. In addition,
-<ul>
-<li>at dataset (aggregate) level, models are
-<ul>
-<li>explainable</li>
-<li>calibrated</li>
-<li>assessed for robustness</li>
-</ul></li>
-<li>at inference time, for every instance, the following are available
-<ul>
-<li>conformal predictions</li>
-<li>explanations</li>
-<li>trust scores</li>
-</ul></li>
-</ul></li>
-</ul>
-<p>Note:</p>
-<ul>
-<li>You can choose any problem or build on the Tabular Data provided in the class. Above must be available on <em>major-project</em> branch of your course private repo.</li>
-<li>Submit your proposal no later than October 20th, 2024, Friday, 13.00pm IST. In the repo, create a <em>README.md</em> file at the root.</li>
-<li>In addition to the code, you have present the project on 26th-29th November, in-class.</li>
-</ul>
-<p>Due:<br>
-</p>
-<p>Proposal Submission:<br>
-- 11.59PM IST, Friday, 20th Oct, 2024.</p>
-<p>Presentations:<br>
-- 26th and 29th, Nov, 2024, in-class.</p>
-<p>Final Submission:<br>
-- 11.59PM IST, Friday, 29th Nov, 2024.</p>
-</section>
-<section id="sec-hw-minor" class="level2">
-<h2 class="anchored" data-anchor-id="sec-hw-minor">Minor Project</h2>
-<p><strong>Task</strong></p>
-<ul>
-<li>This minor (mini) project focuses on baking all the software engineering best practices and MLOps lifecycle into one code base</li>
-<li>Create a branch named <em>minor-project</em> of your private repo.</li>
-<li>For the dataset including all Tranche shared on your google drive, show that, your Models, Data, and Pipelines are
-<ul>
-<li>Project Cards, Data Cards, Model Cards, MLOps Cards are populated and are data driven.</li>
-<li>Code is modular, linted, and tested</li>
-<li>CI/CD hooks are implemented</li>
-<li>Model is deployed and available via an API.</li>
-<li>Inference can be scaled</li>
-<li>Model is monitored, and a new model is deployed when an opportunity or a need to redeploy is detected.</li>
-<li>Right to be forgotten is enabled.</li>
-</ul></li>
-<li>One way to think about it is, it is is culmination of HWs up until now.</li>
-</ul>
-<p><strong>Due by</strong><br>
-11.59PM IST, Friday, 25th Oct, 2024.</p>
-</section>
 <section id="sec-hw-06" class="level2">
 <h2 class="anchored" data-anchor-id="sec-hw-06">HW-06</h2>
 <p><strong>Reading</strong></p>

lectures/w07-l02.html

@@ -348,7 +348,8 @@ <h3 class="anchored" data-anchor-id="additional-reading-optional">Additional Rea
 <ol type="1">
 <li>[paper] <a href="https://arxiv.org/abs/2110.08420">Understanding Dataset difficulty</a></li>
 <li>[tools] <a href="https://github.com/kkirchheim/pytorch-ood">Pytoch-ood</a> - a collection of techniques to detect OOD in PyTroch. Mostly image focussed.</li>
-<li>[tools] <a href="https://github.com/yzhao062/pyod">PyOD</a> - a collection of Anomaly detection techniques</li>
+<li>[tools] <a href="https://github.com/yzhao062/pyod">PyOD</a> - a collection of anomaly detection techniques</li>
+<li>[tools] <a href="https://github.com/deel-ai">DEEL</a> - a collection of OOD, XAI, and other techniques</li>
 </ol>
 </section>
 </section>

search.json

@@ -545,7 +545,7 @@
     "href": "lectures/w07-l02.html#materials",
     "title": "07B: Sample Hardness",
     "section": "",
-    "text": "Pre-work:\n\nAIC a criteria for model selection\ncords for a collection of works/implementations based on subset selection\n\n\n\nIn-Class\n\nCharacterizing data difficulty or sample hardness.\nLook at some statistics like Relative Mahalanobis Distance (which some used to flag OOD, others used to measure sample hardness), Perplexity (cross entropy between two models, one model and data or between), Trust Scores\nSample easiness on training performance and generalization error\n\n\n\nPost-class\n\n[paper] Learning Sample Difficulty from Pre-trained Models for Reliable Prediction\n[paper] A Simple Fix to Mahalanobis Distance for Improving Near-OOD Detection\n[paper] Dissecting Sample Hardness: A fine-grained analysis of hardness characterization methods for data-centric AI\n[paper] To Trust or Not To Trust A Classifier\n\n\n\nAdditional Reading (optional)\n\n[paper] Understanding Dataset difficulty\n[tools] Pytoch-ood - a collection of techniques to detect OOD in PyTroch. Mostly image focussed.\n[tools] PyOD - a collection of Anomaly detection techniques",
+    "text": "Pre-work:\n\nAIC a criteria for model selection\ncords for a collection of works/implementations based on subset selection\n\n\n\nIn-Class\n\nCharacterizing data difficulty or sample hardness.\nLook at some statistics like Relative Mahalanobis Distance (which some used to flag OOD, others used to measure sample hardness), Perplexity (cross entropy between two models, one model and data or between), Trust Scores\nSample easiness on training performance and generalization error\n\n\n\nPost-class\n\n[paper] Learning Sample Difficulty from Pre-trained Models for Reliable Prediction\n[paper] A Simple Fix to Mahalanobis Distance for Improving Near-OOD Detection\n[paper] Dissecting Sample Hardness: A fine-grained analysis of hardness characterization methods for data-centric AI\n[paper] To Trust or Not To Trust A Classifier\n\n\n\nAdditional Reading (optional)\n\n[paper] Understanding Dataset difficulty\n[tools] Pytoch-ood - a collection of techniques to detect OOD in PyTroch. Mostly image focussed.\n[tools] PyOD - a collection of anomaly detection techniques\n[tools] DEEL - a collection of OOD, XAI, and other techniques",
     "crumbs": [
       "Lectures[ML Science]",
       "07B: Sample Hardness"
@@ -567,27 +567,7 @@
     "href": "homeworks.html",
     "title": "Homeworks",
     "section": "",
-    "text": "Major Project\nTask\nNote:\nDue:\nProposal Submission:\n- 11.59PM IST, Friday, 20th Oct, 2024.\nPresentations:\n- 26th and 29th, Nov, 2024, in-class.\nFinal Submission:\n- 11.59PM IST, Friday, 29th Nov, 2024.",
-    "crumbs": [
-      "Homeworks"
-    ]
-  },
-  {
-    "objectID": "homeworks.html#sec-hw-major",
-    "href": "homeworks.html#sec-hw-major",
-    "title": "Homeworks",
-    "section": "",
-    "text": "This major project focuses on baking all the software engineering best practices, MLOps lifecycle and holistic ML into one code base. All the objectives of the Minor Project should be there. In addition,\n\nat dataset (aggregate) level, models are\n\nexplainable\ncalibrated\nassessed for robustness\n\nat inference time, for every instance, the following are available\n\nconformal predictions\nexplanations\ntrust scores\n\n\n\n\n\nYou can choose any problem or build on the Tabular Data provided in the class. Above must be available on major-project branch of your course private repo.\nSubmit your proposal no later than October 20th, 2024, Friday, 13.00pm IST. In the repo, create a README.md file at the root.\nIn addition to the code, you have present the project on 26th-29th November, in-class.",
-    "crumbs": [
-      "Homeworks"
-    ]
-  },
-  {
-    "objectID": "homeworks.html#sec-hw-minor",
-    "href": "homeworks.html#sec-hw-minor",
-    "title": "Homeworks",
-    "section": "Minor Project",
-    "text": "Minor Project\nTask\n\nThis minor (mini) project focuses on baking all the software engineering best practices and MLOps lifecycle into one code base\nCreate a branch named minor-project of your private repo.\nFor the dataset including all Tranche shared on your google drive, show that, your Models, Data, and Pipelines are\n\nProject Cards, Data Cards, Model Cards, MLOps Cards are populated and are data driven.\nCode is modular, linted, and tested\nCI/CD hooks are implemented\nModel is deployed and available via an API.\nInference can be scaled\nModel is monitored, and a new model is deployed when an opportunity or a need to redeploy is detected.\nRight to be forgotten is enabled.\n\nOne way to think about it is, it is is culmination of HWs up until now.\n\nDue by\n11.59PM IST, Friday, 25th Oct, 2024.",
+    "text": "HW-06\nReading\nTask\nDue by\n11.59PM IST, Friday, 18th Oct, 2024.",
     "crumbs": [
       "Homeworks"
     ]
@@ -596,8 +576,8 @@
     "objectID": "homeworks.html#sec-hw-06",
     "href": "homeworks.html#sec-hw-06",
     "title": "Homeworks",
-    "section": "HW-06",
-    "text": "HW-06\nReading\n\nRight to Erasure act of GDPR\n\nTask\n\nThis HW focuses on the ability to implement techniques to enforce the “right to erase” policy\nCreate a branch named hw06 of your private repo.\nTreat the model you have built in HW-04 as your baseline, call it Model A\nImagine the first 10 records were to be erased.\nImplement a strategy s.t those 10 records will not be used in any future predictions. Update the API, and Data and Model cards to reflect this change.\nSuggest a method to implement “right to erasure” such that, even all past predictions that were influenced by the 10 records will be re-scored, and pushed.\n\nDue by\n11.59PM IST, Friday, 18th Oct, 2024.",
+    "section": "",
+    "text": "Right to Erasure act of GDPR\n\n\n\nThis HW focuses on the ability to implement techniques to enforce the “right to erase” policy\nCreate a branch named hw06 of your private repo.\nTreat the model you have built in HW-04 as your baseline, call it Model A\nImagine the first 10 records were to be erased.\nImplement a strategy s.t those 10 records will not be used in any future predictions. Update the API, and Data and Model cards to reflect this change.\nSuggest a method to implement “right to erasure” such that, even all past predictions that were influenced by the 10 records will be re-scored, and pushed.",
     "crumbs": [
       "Homeworks"
     ]

sitemap.xml

@@ -66,19 +66,19 @@
   </url>
   <url>
     <loc>https://mlsquare.github.io/ai-839/lectures/w07-l02.html</loc>
-    <lastmod>2024-09-12T10:00:46.191Z</lastmod>
+    <lastmod>2024-09-12T18:07:38.616Z</lastmod>
   </url>
   <url>
     <loc>https://mlsquare.github.io/ai-839/homeworks.html</loc>
-    <lastmod>2024-09-05T07:17:03.927Z</lastmod>
+    <lastmod>2024-09-12T10:59:23.996Z</lastmod>
   </url>
   <url>
     <loc>https://mlsquare.github.io/ai-839/projects.html</loc>
     <lastmod>2024-09-05T10:10:35.880Z</lastmod>
   </url>
   <url>
     <loc>https://mlsquare.github.io/ai-839/tutorials.html</loc>
-    <lastmod>2024-08-25T06:48:04.143Z</lastmod>
+    <lastmod>2024-09-12T10:59:37.247Z</lastmod>
   </url>
   <url>
     <loc>https://mlsquare.github.io/ai-839/talks.html</loc>