https://sanjaykthakur.com/more-about-me/2024-03-23T10:22:39+00:00weekly0.6https://sanjaykthakur.com/2022/04/18/scene-graphs-for-driving-scenario-description/2022-04-18T12:07:35+00:00monthlyhttps://sanjaykthakur.com/2022/03/24/the-long-tail-of-rare-but-realistic-scenarios-are-the-true-thorn-in-the-av-adas-softwares-flesh/2022-03-24T11:19:15+00:00monthlyhttps://sanjaykthakur.com/2022/03/17/measuring-scalability-for-autonomous-mobility-solutions-and-supervised-ml/https://sanjaykthakur.com/wp-content/uploads/2022/03/1kfzyfut2okfjekjlceyvuq.jpeg1kfzyfut2okfjekjlceyvuq2022-03-17T11:55:03+00:00monthlyhttps://sanjaykthakur.com/2021/01/18/launching-rydesafely/2021-09-09T12:21:57+00:00monthlyhttps://sanjaykthakur.com/2021/01/30/my-main-learning-from-3-month-long-power-packed-ef-form/2021-06-18T18:59:11+00:00monthlyhttps://sanjaykthakur.com/2021/06/11/light-does-not-travel-at-the-same-speed-in-all-directions-or-does-it/2021-06-17T14:45:08+00:00monthlyhttps://sanjaykthakur.com/2021/04/09/positioning-machine-learning-for-emerging-india-and-developing-world-last-updated-in-2019/2021-04-10T01:29:09+00:00monthlyhttps://sanjaykthakur.com/2020/07/26/the-legacy-of-sushant-singh-rajput/2020-07-27T12:03:36+00:00monthlyhttps://sanjaykthakur.com/2015/10/10/uqam-montrealmy-first-international-work-experience/https://sanjaykthakur.com/wp-content/uploads/2015/10/ssmp7.jpegThe Bixi bike I used to use2020-07-26T18:56:40+00:00monthlyhttps://sanjaykthakur.com/2016/05/28/the-key-to-win-depends-the-extra-mile-you-take/https://sanjaykthakur.com/wp-content/uploads/2016/05/22838692697_eda9676c43_o.jpg22838692697_eda9676c43_o2020-07-26T18:56:27+00:00monthlyhttps://sanjaykthakur.com/2017/10/09/30-things-i-want-to-do-before-i-die/2020-07-26T18:55:43+00:00monthlyhttps://sanjaykthakur.com/2020/07/03/a-new-economy-called-space-and-its-transformative-effects-on-earth/2020-07-09T08:14:02+00:00monthlyhttps://sanjaykthakur.com/2019/11/08/hardware-based-challenges-for-building-autonomous-robots/https://sanjaykthakur.com/wp-content/uploads/2019/04/take-this.pngtake this2019-11-14T10:49:54+00:00monthlyhttps://sanjaykthakur.com/2019/10/29/challenges-for-building-neural-network-based-autonomous-robots/2019-11-08T14:46:15+00:00monthlyhttps://sanjaykthakur.com/2019/03/11/uncertainty-aware-learning-from-demonstrations-in-multiple-contexts-using-bayesian-neural-networks/https://sanjaykthakur.com/wp-content/uploads/2019/03/pendulum.pngpendulumhttps://sanjaykthakur.com/wp-content/uploads/2019/03/real_context_6_to_real-1.pngreal_context_6_to_realhttps://sanjaykthakur.com/wp-content/uploads/2019/03/visualization-1.pngvisualization (1)https://sanjaykthakur.com/wp-content/uploads/2019/03/3.png3https://sanjaykthakur.com/wp-content/uploads/2019/03/2-1.png2https://sanjaykthakur.com/wp-content/uploads/2019/03/our_robot_pendulum.jpgour_robot_pendulum2019-05-16T14:25:56+00:00monthlyhttps://sanjaykthakur.com/2018/12/23/learning-from-demonstrations-i/https://sanjaykthakur.com/wp-content/uploads/2018/12/DAGGER.pngDAGGER\emph{DAGGER}~(\cite{Dagger}) is an iterative and interactive LfD technique that has demonstrated impressive results on learning policies by directly mimicking the demonstrations. The red boxes highlights the important steps in this method as described in figure~\ref{fig:interactive_imitation_learning}.https://sanjaykthakur.com/wp-content/uploads/2018/12/Interactive_LfD.pngInteractive_LfDAny interactive LfD technique repetitively keeps training itself on the demonstration feedback received on the learner rollouts.https://sanjaykthakur.com/wp-content/uploads/2018/12/BC_vs_Interactive_BC.pngBC_vs_Interactive_BCNaive Behavioral Cloning (left figure) leads to gradual away drift from the demonstrations that can lead to potentially unwanted situation. Interactive approaches (right figure) mitigates this problem by having learner use demonstrator feedback on learner induced states. Image used from \href{http://rll.berkeley.edu/deeprlcourse-fa15/docs/2015.10.5.dagger.pdf}{here}.https://sanjaykthakur.com/wp-content/uploads/2018/12/BC_Goal.pngBC_Goalhttps://sanjaykthakur.com/wp-content/uploads/2018/12/torcs.jpgtorcsA snapshot of the TORCS simulator~(\cite{wymann2000torcs}). Here $s$ would be the game screen and $a$ would be the amount of throttle and steering angles of the car. $\pi^*$ can be a human demonstrator controlling a car using a joystick and $\pi_{\theta}$ can be a neural network policy. $\tau$ would be the coupled state-action pairs at every time-step in a game run and the loss (surrogate) function can be mean-squared error on the demonstrated and predicted actions.https://sanjaykthakur.com/wp-content/uploads/2018/12/LfD_goal.pngLfD_goalhttps://sanjaykthakur.com/wp-content/uploads/2018/12/FosburyFlop.jpgFosburyFlopFosbury Flop: Until demonstrated by Dick Fosbury in 1968 Summer Olympics, all elite high jumpers used lesser effective techniques of high jump. This turn of events has remarkable similarity to finding the optimal policy for problems by current techniques versus by using demonstrations.https://sanjaykthakur.com/wp-content/uploads/2018/12/Human_motivated_LfD.pngHuman_motivated_LfDA bulk of our behavior patterns are motivated by observing others (top figure). Hence in order to attain more capable AI systems, we have to endow them with the mechanism to learn from demonstrations (bottom figure, taken from2019-04-02T01:50:37+00:00monthlyhttps://sanjaykthakur.com/2018/12/05/the-very-basics-of-bayesian-neural-networks/https://sanjaykthakur.com/wp-content/uploads/2018/12/bayesian_statistics.jpgbayesian_statisticsBayesian inference adjusts the beliefs about a distribution in the light of data or evidencehttps://sanjaykthakur.com/wp-content/uploads/2018/12/bayes_rule.pngBayes_Rulehttps://sanjaykthakur.com/wp-content/uploads/2018/12/extrapolation_graph.pngextrapolation_graphhttps://sanjaykthakur.com/wp-content/uploads/2018/12/bayes_nn.pngbayes_NNUnlike deterministic Neural Networks(left) that have a fixed value of their parameters, Bayesian Neural Networks(right) have a distribution defined over them2019-04-02T01:50:12+00:00monthlyhttps://sanjaykthakur.com/about/transferability-and-sample-efficiency-in-data-driven-ai/2019-03-05T21:50:05+00:00weekly0.6https://sanjaykthakur.com/2017/05/25/installing-experience-into-artificial-intelligent-agents/https://sanjaykthakur.com/wp-content/uploads/2017/05/still-waiting.jpgStill WaitingOur superhuman AI algorithms are data hungryhttps://sanjaykthakur.com/wp-content/uploads/2017/05/pong.gifPongPong, taken from http://uncyclopedia.wikia.com/wiki/File:Exciting_Pong.gifhttps://sanjaykthakur.com/wp-content/uploads/2017/05/lstm.jpgLSTMLSTM, Taken from : http://colah.github.io/posts/2015-08-Understanding-LSTMs/https://sanjaykthakur.com/wp-content/uploads/2017/05/breakout.gifBreakoutBreakout, taken from https://linustechtips.com/main/topic/317702-googles-new-ai-has-already-learnt-how-to-crush-us-at-49-games-update/2018-12-06T22:18:01+00:00monthlyhttps://sanjaykthakur.com/2016/01/10/detection-of-antipatterns-in-mobile-architectures-and-resolving-them/2018-05-08T23:06:08+00:00monthlyhttps://sanjaykthakur.com/2016/11/11/rendezvous-with-realistic-image-synthesis/https://sanjaykthakur.com/wp-content/uploads/2016/11/7.jpg7Another rendered image by ray tracing. Source: http://blog.zao3d.com/rendering-synthetic-objects-into-legacy-photographs/https://sanjaykthakur.com/wp-content/uploads/2016/11/8.png8A rendered image by ray tracing. Source: Wikipediahttps://sanjaykthakur.com/wp-content/uploads/2016/11/9.jpg9A rendered image. Source: http://blog.zao3d.com/rendering-synthetic-objects-into-legacy-photographs/https://sanjaykthakur.com/wp-content/uploads/2016/11/6.jpg6https://sanjaykthakur.com/wp-content/uploads/2016/11/2.jpg2A rendered image for advertisement. Source: http://www.honda.ca/https://sanjaykthakur.com/wp-content/uploads/2016/11/3.jpg3A rendered image for advertisement. Source: http://www.honda.ca/https://sanjaykthakur.com/wp-content/uploads/2016/11/4.jpg4A rendered image for advertisement. Source: http://icdn-7.motor1.com/images/mgl/PvbZK/s1/bmw-m1-sedan-rendering.jpghttps://sanjaykthakur.com/wp-content/uploads/2016/11/5.jpg5A rendered image. Source: http://www.digitaltutors.com/lesson/18052-Compositing-rendered-images-and-the-conclusionhttps://sanjaykthakur.com/wp-content/uploads/2016/11/shaderlight-rendering-4.jpgshaderlight-rendering-4A rendered image. Source: http://www.pxleyes.com/blog/2010/03/5-things-you-need-to-know-about-raytracing/2018-05-08T23:05:46+00:00monthlyhttps://sanjaykthakur.com/2017/06/17/interfacing-tensorflow-to-noobs/https://sanjaykthakur.com/wp-content/uploads/2017/06/trends.pngtrendshttps://sanjaykthakur.com/wp-content/uploads/2017/06/tensorflow-tutorial1.pngTensorFlow Tutorialhttps://sanjaykthakur.com/wp-content/uploads/2017/06/diagram1.jpegDiagram1https://sanjaykthakur.com/wp-content/uploads/2017/06/optimize.jpegOptimizehttps://sanjaykthakur.com/wp-content/uploads/2017/06/loss.jpegLosshttps://sanjaykthakur.com/wp-content/uploads/2017/06/infer.jpegInferhttps://sanjaykthakur.com/wp-content/uploads/2017/06/quiz-1.jpegQuiz 1https://sanjaykthakur.com/wp-content/uploads/2017/06/steps.jpgStepshttps://sanjaykthakur.com/wp-content/uploads/2017/06/fig-3-sample-images-of-mnist-data.pngFig-3-Sample-images-of-MNIST-dataTaken from https://www.researchgate.net/publication/222834590_Handwritten_digit_recognition_Benchmarking_of_state-of-the-art_techniqueshttps://sanjaykthakur.com/wp-content/uploads/2017/06/fig-18-0-9-sample-digits-of-mnist-handwritten-digit-database.pngFig-18-0-9-Sample-digits-of-MNIST-handwritten-digit-databaseTaken from https://www.researchgate.net/publication/222834590_Handwritten_digit_recognition_Benchmarking_of_state-of-the-art_techniques2018-05-08T23:01:58+00:00monthlyhttps://sanjaykthakur.comdaily1.02024-03-23T10:22:39+00:00