ദേവാനന്ദ് | Devanand

I acquired ~700 historical documents from Saleem al-Rayes, Gaza's #1 antique dealer. The documents span the periods, 1910s-1970s. You can download the docs for free here: palestinenexus.com/research Here is video of Saleem explaining why he loves Gaza so much, shot on Oct. 1, 2023:

Original Visuals of Dr. Ambedkar. Thanks Babasaheb for Everything, Jai Bhim 🙏

@Israel @Liorfink Lol. Half of the security cabinet have been regurgitating genocidal rhetoric towards Lebanon. By “peace?” — asked by the Israeli — you mean “annihilation”?

@donyadelsouz Zionist derangement syndrome: the belief that carrying out wars of aggressions, ethnic cleansing and genocide make Israel the most moral country in the world.

@SuryaGanguli Blockchain?

Just wanted to point out that recent geopolitical events provide a pedagogical example for teaching the max-flow min-cut theorem.

Viktor Orbán’s huge defeat in the Hungarian election creates hope for all progressive forces in the world. The defeat suggests that ordinary Hungarians realized that right-wing ideology is nothing but a cover for cronyism.

i scream you scream we all scream for ethnic cleansing and genocide

Nithin Raj, a 23 year old BDS student at Anjarakandy Dental College, Kannur died by suicide after relentless caste based abuse from seniors & faculty, including HOD Dr M K Ram & Prof Sangeetha. Another life lost to caste, and some still have the nerve to pretend it doesn’t exist!

A monumental portrait of Dr.B.R.Ambedkar using 18000 notebooks.🔥😍 By artist Chetan Raut. #jaibhim

it's genocide ( @amnesty ) it's genocide ( @hrw ) it's genocide ( @btselem ) it's genocide ( @MSF ) it's genocide (IAGS) it's genocide ( @alhaq_org ) it's genocide ( @UNHumanRights ) it's genocide ( @UN_HRC ) it's genocide ( @pchrgaza ) it's genocide ( @AlMezanCenter ) it's genocide ( @WarOnWant ) it's genocide ( @PHRIsrael ) it's genocide ( @fidh_en ) it's genocide (PHROC) it's genocide ( @LemkinInstitute ) it's genocide ( @theCCR ) it's genocide ( @ECCHRBerlin ) it's genocide ( @unitedforrights ) it's genocide ( @JURDIasso ) it's genocide ( @TheElders ) it's genocide ( @Oxfam ) calling a genocide a genocide hurts my feelings (Erielle Azerrad)

Excited to share my latest work: Teacher-Free Self-Distillation (TFSD)! A new loss function that fixes deep flaws in standard cross-entropy for classification. No more infinite gaps or unstable training. Let's dive in. 🧵 1/8

Neural networks have a fundamental problem. Feed them garbage data and instead of admitting that they are confused, they will confidently hallucinate. I just open-sourced the HALO-Loss to try and fix this. It give the model a mathematically sound *I don't know!* button.🧵

@slimer48484 @sameQCU I can share the code. Give me some time. Can I DM you?

@devanand_t @sameQCU Did you initialize it in a specific way to make it work? Would I be able to read your code?

Geoffrey Hinton did an MNIST subliminal learning experiment in the 2015 paper 'Distilling the Knowledge in a Neural Network'! They trained a neural network on a dataset that completly lacked the number 3 and it exactly what the number 3 looked like anyway. What's even crazier is that you can train a neural network entirely on nosie data with ghost logits that dont correspond to any real numbers and have it learn to classify all the numbers!

@slimer48484 @sameQCU You mean replicating student from ghost data? I have done a Resnet18 experiment with CIFAR10 for teacher.

@sameQCU weirdly while it works for MLPs but I'm not getting it happening with CNNs (yet?)

I talked to two Israeli sources on why Iranian launches continue to increase, despite US-Israeli claims that they have destroyed almost all of the launchers. Here is what they said: 1) The 90–95% drop in volume claimed by CENTCOM earlier in the month was probably a temporary lull as Iran repositioned its remaining launchers into hardened sites. Independent satellite analysis suggests that a significant portion of the "80% destruction" claimed by the IDF actually hit high-fidelity decoys. 2) Despite fewer launchers, the lethality per strike has increased. Iran's shift to cluster warheads has allowed a single missile to impact multiple locations simultaneously, compensating for the lower volume of launches 3) Iran has successfully set up mobile, underground units able to fire at steady rates. Iran used that quiet period to move their remaining ~100-120 heavy launchers into "Super-Hardened" facilities 4) Iran is utilizing its Zolfaqar and Dezful road-mobile launchers. These units move from hardened tunnels to pre-surveyed launch spots, fire, and return underground in under 10 minutes, often before coalition drones can re-task for a strike. 5) Because these launching units are decentralized, it is very hard for US and Israeli intelligence to get info on them. Israel and the United States do not have an answer to this problem. That is why they are trying escalation on energy sources instead. But that is backfiring.

We introduce the Perceptual Manifold (PM): the set of all inputs a network confidently labels as a given class. Our foundational finding: machine PMs are orders of magnitude higher dimensional than human ones.

Holy shit... Microsoft open sourced an inference framework that runs a 100B parameter LLM on a single CPU. It's called BitNet. And it does what was supposed to be impossible. No GPU. No cloud. No $10K hardware setup. Just your laptop running a 100-billion parameter model at human reading speed. Here's how it works: Every other LLM stores weights in 32-bit or 16-bit floats. BitNet uses 1.58 bits. Weights are ternary just -1, 0, or +1. That's it. No floats. No expensive matrix math. Pure integer operations your CPU was already built for. The result: - 100B model runs on a single CPU at 5-7 tokens/second - 2.37x to 6.17x faster than llama.cpp on x86 - 82% lower energy consumption on x86 CPUs - 1.37x to 5.07x speedup on ARM (your MacBook) - Memory drops by 16-32x vs full-precision models The wildest part: Accuracy barely moves. BitNet b1.58 2B4T their flagship model was trained on 4 trillion tokens and benchmarks competitively against full-precision models of the same size. The quantization isn't destroying quality. It's just removing the bloat. What this actually means: - Run AI completely offline. Your data never leaves your machine - Deploy LLMs on phones, IoT devices, edge hardware - No more cloud API bills for inference - AI in regions with no reliable internet The model supports ARM and x86. Works on your MacBook, your Linux box, your Windows machine. 27.4K GitHub stars. 2.2K forks. Built by Microsoft Research. 100% Open Source. MIT License.

If you're using Claude Code for research: stop making it read directly from PDFs We've introduced a SKILL.md that fetches structured, AI-friendly paper overviews from alphaXiv 👀

We’re excited to release TorchLean which is the first fully verified neural network framework in Lean. The Lean community has largely focused on pure mathematics. TorchLean expands this frontier toward verified neural network software and scientific computing. With the recent release of CSlib, we see this as another step toward a fully verified ML stack. We support features: 1. Executable IEEE-754 floating-point semantics (and extensible alternative FP models) verified tensor abstractions with precise shape/indexing semantics 2. Formally verified autograd system for differentiation of NN programs Proof-checked certification / verification algorithms like CROWN (robustness, bounds, etc.) 3. PyTorch-inspired modeling API with eager-style development + export/lowering to a shared IR for execution and verification Project page: leandojo.org/torchlean.html Paper: [2602.22631] TorchLean: Formalizing Neural Networks in Lean Work done @Robertljg, Jennifer Cruden, Xiangru Zhong, @huan_zhang12 and @AnimaAnandkumar. #MachineLearning #ScientificComputing #Lean