Loading…
Topic
615 posts about Machine Learning. Every summary links to the original.
See the idea
A walkthrough of how a 3D CT model turns a voxel volume into a structured diagnostic report: isotropic resampling, 3D patch tokens, spatial encoding, and language-model reasoning.
The Good Stuff
Paid picks we use ourselves. Links carry a referral code — a subscription through one earns us a commission at no extra cost to you.
Newest first
An ICML 2026 reproduction hackathon examined whether coding agents could audit research at conference scale, amid 23,918 submissions and 6,352 accepted papers. From July 15 to August 2, 1,221 participants published 6,816 Trackio logbooks covering 2,226 papers and 35,908 claims; an open-weights GLM-5.2 judge assigned per-claim verdicts, while participants used cloud jobs and toy reproductions when full experiments were impossible. At least one claim was independently verified in 51% of examined papers, including 266 fully reproduced papers, while 23% had a falsified or contested claim and 242 papers produced opposing verdicts across teams. Adversarial re-checks confirmed issues including a paging-theorem error, a reverse-versus-forward-KL mismatch, and padded evaluation tokens, but also caught false falsifications, leading to the conclusion that agents expand auditing capacity while humans remain necessary to steer experiments and assess cases such as image quality.
OlmoEarth Studio now supports custom computation and export of embedding vectors, compact representations of Earth-observation data generated by open-source OlmoEarth foundation models. Users can select an area, a one-to-twelve-month time span, an encoder variant, spatial resolution, and Sentinel-2 L2A or Sentinel-1 RTC imagery, then receive a Cloud-Optimized GeoTIFF with one int8 band per embedding dimension. The post demonstrates similarity search, few-shot segmentation, change detection, and PCA-based exploration; a logistic regression trained on 60 labeled pixels produced a weighted F1 of 0.84 for three land-cover classes in Ca Mau, Vietnam. Embeddings provide a fast, shareable starting point, while supervised fine-tuning can improve performance for applications needing task-specific results; cloud cover, atmospheric artifacts, and missing observations can affect quality.
WhatsApp presents an early technical overview of Scam Alert, an optional feature that uses an on-device machine learning model to identify potential scam messages from non-contacts while keeping message content on the device. The warning appears only to the user, who can block, report, continue, or mark a chat trusted; reporting and sharing require explicit user action. To measure performance, the system sends locally aggregated warning and user action counts through confidential federated analytics using Trusted Execution Environments and differential privacy, rather than transmitting raw signals. Model versions are recorded on a public transparency ledger, while published weights, client-side logs, and an expanded Bug Bounty program support independent inspection of delivery and behavior.
NVIDIA Nemotron 3.5 Lightning is an open 30B mixture-of-experts model with 3B active parameters, designed for the high-volume execution layer of long-running autonomous agents. The post positions it alongside frontier reasoning models: larger systems handle planning and orchestration, while Lightning handles tool calls, result validation, subagent delegation, and other routine work. Its design combines harness-optimized training, multi-token prediction for speculative decoding, DFlash and DSpark draft models, and NVFP4 quantization alongside BF16. The model is customizable through LoRA or full SFT, reinforcement learning, and environment-based evaluations, with weights, training data, and recipes released under OpenMDW-1.1. The post reports up to 4x the output speed of similar-sized models, 86% accuracy on PinchBench, and completion of 10,000 tasks 30% faster than Qwen3.6 35B at similar accuracy, while NeMo Switchyard routes work among models.
amazon.science
Modern language model architectures evolved around specific hardware constraints, but AWS Trainium chips alter memory and compute dynamics through on-chip SBUF memory, systolic matrix multiplication, and explicit DMA control. The AWS Trainium Frontier competition invites teams to train language models from scratch under strict wall-clock time constraints to uncover optimal architectures and custom kernels for Trainium silicon. Participants start from a fifty-million-parameter dense baseline and can modify everything from the model architecture and optimizer to custom Neuron Kernel Interface routines. Phase 1 evaluates thirty-minute single-chip training runs by validation bits-per-byte, while Phase 2 grants four-hour multi-chip server access to evaluate both training efficiency and reasoning on the CORE benchmark. Top finalists will present their findings at NeurIPS 2026, compete for cash prizes, and co-publish research with Annapurna Labs.
NVIDIA Magpie Multilingual TTS is a 364M-parameter open-weights model for multilingual voice applications, supporting 12 languages, including newly added Modern Standard Arabic, Korean, and Brazilian Portuguese. It is positioned for cascaded voice-agent architectures in which ASR, TTS, and LLM components are independently deployed, tuned, and scaled on controlled infrastructure. Frame stacking predicts two audio frames per decoding step, while a local transformer models dependencies between simultaneously generated codebook tokens; NVIDIA reports 32–79 ms single-stream TTFA across its listed GPUs. Compared with the previous release, French and Spanish CER improved to 1.54% and 0.60%, while SSIM reached 0.747 and 0.793; German SSIM also increased to 0.742. The Hugging Face checkpoint supports research and fine-tuning, while NVIDIA NIM provides optimized production serving and NeMo supports customization.
Knowledge distillation trains a smaller student to match a larger teacher's capabilities, but conventional online training keeps both models loaded and materializes full vocabulary distributions at every step, making long-context recovery expensive. The proposed system caches the teacher's top-100 logits once and introduces a fused, chunked KL-divergence loss that avoids storing the full vocabulary-by-sequence grid or the student's full logits. In benchmarks with an 8K-token context on one H200, offline fused chunked KL used 58.3GB peak memory versus 102.8GB for online training, while all four methods reached near-identical training loss. At 32K tokens, the fully chunked loss reduced isolated kernel memory from 85.2 GiB to 5.45 GiB; distilling GPT-OSS 20B at that context cut step time from 57.0 to 12.23 seconds and reduced the setup from four GPU nodes to one.
Web traffic increasingly shifts between human interaction and automated agents, rendering point-in-time security checks ineffective against sophisticated automation. Cloudflare addresses this shift through continuous behavioral analysis designed to evaluate trust across entire user sessions rather than relying on static point-in-time hurdles. Its CDN-injected client-side system, Precursor, tracks continuous behavioral indicators like cursor movements to catch mid-session automated takeovers. To counter rapid bot adaptation, Cloudflare is introducing self-adjusting machine learning models under Adaptive Intelligence and non-deterministic defenses such as the AI Labyrinth. These tools allow site owners to poison scraper training data, trap crawlers in synthetic webs, or queue legitimate agents while dynamically penalizing abusive traffic.
Meta’s ads ranking systems process millions of candidates per second under millisecond latency budgets, creating tension between richer sequence models and efficient serving. The post presents a multi-stage architecture that moves long-history user modeling into asynchronous offline transformers, caches user-level embeddings, and combines them with fresh user and ad signals in an online ranking model. Dense tokenization places sparse features and behavioral sequences in one vocabulary, while target-aware multi-head attention learns higher-order interactions between historical behavior and the candidate ad. On real-world ads traffic, performance follows a log-linear relationship with compute across depth, enrichment, width, and sequence length, with balanced growth and diverse sequences identified as scaling levers. Together with broader model innovations, the approach contributed to a cumulative 6% conversion lift on Instagram, 3% on Facebook, and 3.5% lift in Facebook ad clicks, and serves as a core component of GEM.
amazon.science
Amazon announced 34 recipients from 30 universities for the Fall 2025 Build on Trainium: Responsible AI call for proposals. Part of a 110 million dollar credit initiative, the program funds academic research across new model architectures, machine learning libraries, optimizations, and distributed systems running on AWS Trainium hardware. The funded projects center on five key areas: AI safety and alignment, multilingual language models, representation engineering, sustainability with small language models, and deep learning for synthetic data generation. Awardees receive AWS Promotional Credits, access to more than 700 Amazon public datasets, assigned Amazon research contacts, and dedicated technical training resources. Highlighted efforts include parallelization strategies for trillion-parameter mixture-of-experts models scaling across 1,024 Trainium chips and inference optimization frameworks designed to enhance token efficiency.
Frontier models can accelerate an AI product launch, but their serving cost and latency can become impractical as usage grows, while their weights do not learn from production failures. Shopify’s proposed flywheel captures low-quality conversations, defines quality with a rubric and calibrated judges, improves the surrounding harness through autoresearch, and then converts repaired trajectories into supervised fine-tuning and GRPO training signal. A panel of frontier reasoning models critiques failures, an arbiter produces repair instructions, and unresolved cases go to expert annotators; gist compression then shortens the agent’s static prompt without measured judge-quality loss. In production, the GraphQL agent serves up to 2,000 requests per minute and reportedly achieves a 96% serving-cost reduction versus an estimated frontier-model cost, with lower latency and higher throughput. The approach is presented as a continual-learning system that turns production experience into model-weight updates.
Robot policies built on vision-language-action models often struggle with physical generalization because vision-language backbones describe scenes instead of predicting physical dynamics. World action models address this limitation by building policies on video world models that jointly predict actions and future world states. NVIDIA Cosmos 3 provides an open foundation using a Mixture-of-Transformers architecture where an autoregressive transformer guides a diffusion transformer to synthesize continuous modalities including video and robot actions. Specializing Cosmos 3 on the DROID platform produced policies capable of simultaneously outputting action trajectories and predicted video observations while boosting RoboLab success from 28.1% to 36.8%. These models support flexible deployment tiers, from workstation serving on RTX PRO 6000 GPUs to real-time 15 Hz on-device execution on NVIDIA Jetson Thor embedded hardware.
Autonomous vehicle development typically relies on disjoint models for trajectory generation, intent prediction, scene understanding, and data labeling. NVIDIA Alpamayo 2 Super addresses this fragmentation with an open 34-billion-parameter vision-language-action model that couples a 32-billion-parameter Cosmos 3 Super Reasoner with a 2-billion-parameter diffusion-based Action Expert. Processing surround-view video across up to seven cameras, the model simultaneously produces future vehicle trajectories, Chain-of-Causation reasoning traces, high-level meta-actions, and grounded scene question answers. In benchmark evaluations, the model achieved a 6.4-second minADE_6 of 0.911 meters on the Physical AI AV Dataset and ranked first on LingoQA with a score of 79.2. Released under the OpenMDW-1.1 permissive license, the architecture serves as an offline policy teacher, evaluation critic, or auto-labeling engine that can be distilled for NVIDIA DRIVE AGX Thor deployment.
Organizations sharing a single Kubernetes cluster across multiple teams frequently encounter conflicting custom resource definitions, overlapping role-based access control, and difficulties budgeting GPU capacity. To address these operational conflicts without provisioning separate hardware, administrators can deploy vCluster alongside KAI Scheduler on a shared host cluster. vCluster supplies each team with a dedicated, virtualized control plane featuring its own API server and permissions, while KAI Scheduler dynamically time-slices GPU resources through hierarchical queues. In a demonstration using an NVIDIA L40S GPU on MicroK8s, three separate tenant clusters successfully scheduled workloads with fractional allocations of 0.33 GPU each. The combination achieves tenant autonomy and visibility isolation while consolidating AI workloads on shared physical infrastructure.
Agentic AI architectures require storage systems to secure, compress, and validate massive streams of persistent data and key-value cache entries across concurrent reasoning loops. These tasks run primarily on CPUs and frequently create processing bottlenecks that prevent modern networks and flash storage from operating at full capacity. NVIDIA evaluated the Vera CPU within the BlueField-4 STX Storage Processor using a purpose-built in-memory microbenchmark suite running standard libraries like OpenSSL, Zstandard, and LZ4. The results demonstrated that Vera outperformed x86 processors across core storage primitives including encryption, decompression, Reed-Solomon recovery, and CRC32C integrity checking. For a sequential multi-stage pipeline executing compression followed by encryption, Vera achieved up to 3.21x higher throughput than the comparison x86 CPU.
Andy Pavlo announces that he is joining ClickHouse to establish and lead ClickHouse Labs, an industry research team focused on database systems. He describes ClickHouse’s early use of C++ and vectorized query execution with SIMD in 2016, and says its development has remained relevant to his academic research. Labs will work directly with ClickHouse engineers, customers, collaborators, and industry partners rather than operating separately, while also collaborating with the PostgreSQL team on performance and reliability for its managed service. An immediate priority is validating and pushing an existing backlog of ideas and optimizations into production, followed by research into hardware, algorithms, data structures, execution strategies, and DBMS development for AI and agentic workloads. The stated aim is to combine scientific research with technology that matters to users.
As context lengths increase in long-context and agentic workloads, attention increasingly dominates GPU inference runtime. Dense attention performance depends on architectural parameters like group size, head dimension, and sequence length, alongside GPU parallelization strategies. Prefill remains compute-bound and scales quadratically with input sequence length, showing minimal runtime sensitivity to group size variations. Conversely, decode generation is memory-bound and scales linearly with KV cache length, where doubling group size roughly halves decode execution time. To maximize throughput and interactivity on NVIDIA hardware, model architects should push group sizes higher, select head dimensions of 128 or 256, reduce effective KV state, and keep tensor parallelism less than or equal to the number of KV heads.
Netflix developed GenRec, an LLM-backed recommendation ranker built by adapting an internal foundation model for large-scale personalization. Traditional recommendation stacks rely on thousands of hand-crafted features and specialized architectures that are costly to maintain across diverse content types. To replace manual feature pipelines, GenRec verbalizes user histories, metadata, and contexts into natural-language prompts and trains with multi-objective losses, including catalog-aware ranking and reward-weighted alignment. At inference time, the model executes in prefill-only mode on vLLM without decoding text. In large-scale online A/B testing against a mature production ranker, GenRec achieved statistically significant improvements in short-term and long-term metrics while using fewer labeled examples.
Modern discussions surrounding agentic AI increasingly center on whether autonomous systems can be trusted and securely placed at the core of enterprise workflows. While intelligence is generated by underlying models, operational trust depends on the surrounding execution runtime, identity frameworks, governance models, and security boundaries. To address these challenges across an open ecosystem, Docker has joined NVIDIA's Open Secure AI Alliance. This initiative aims to ensure developers can seamlessly route tasks between open-weight and frontier models without rewriting their application architectures or sacrificing control. By extending security and governance into the agent execution environment, the alliance seeks to deliver predictable control as AI agents become fundamental components of software stacks.
amazon.science
Multitask machine learning often suffers from gradient conflicts, drift, and drought when blending multiple objective gradients in single update steps. To address this in graph self-supervised learning, researchers developed ControlG, a framework that allocates discrete computational blocks to individual objectives over time. ControlG operates across three coupled loops: sensing objective difficulty via spectral demand and gradient interference, planning target compute allocations using log-hypervolume Pareto sensitivities, and executing the schedule via a proportional-integral-derivative controller. Evaluated across homophilic and heterophilic graph benchmarks, the approach outperforms traditional multitask gradient-manipulation baselines while incurring low computational overhead. The framework also produces interpretable training timelines and deficit traces.
The best engineering reads, once a week.
