Loading…
Topic
622 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
Data teams frequently field questions about future sales, metric movements, and anomalous data, but answering them typically requires data science teams to build and manage models, creating a bottleneck. Predictive AI Functions, now in Beta, remove that bottleneck by providing SQL-native functions for forecasting, driver analysis, and anomaly detection, backed by state-of-the-art ML models maintained centrally by Databricks. The launch includes ai_forecast for time series projection, ai_top_drivers for segment driver analysis, ai_detect_anomalies for outlier detection, and ai_predict_value and ai_predict_class for regression and classification. These functions can be called directly in SQL, invoked through Genie with plain-language questions, or embedded in dashboards, delivering consistent, high-quality answers without infrastructure management. Benchmarking on fev-bench shows ai_forecast outperforms forecasting functions from leading cloud data warehouses, achieving industry-leading accuracy out of the box with no tuning required.
HiRO-ACEv2 generates global 3 km climate data for extreme weather at roughly 1/300th the cost of the X-SHiELD model, with even larger savings for single-region runs. The system pairs a coarse 100 km AI model (ACE) with the new HiROv2 downscaler that simultaneously predicts rain, wind, and sea-level pressure instead of just rainfall, and has improved skill at reproducing extreme events. When fed coarsened X-SHiELD data HiROv2 matches the target’s high-resolution statistics within 3% for events as rare as one in a million data points, and accurately estimates tropical cyclone energy for storms never seen during training. Over a 10-year simulation the full system’s hurricane energy and extreme event snapshots compare well with X-SHiELD. However, HiRO cannot yet handle other climates because it depends on X-SHiELD-specific ACE tuning; future work includes training on warmer-world datasets and adding temperature and snow outputs.
The post argues that machine learning does not always require advanced mathematics and, as described by Intercom’s Fergal Reid, is about systems that improve with more training data. It contrasts a naive keyword-matching approach in Ruby for a ‘You might also like’ feature, which produced irrelevant results, with a basic machine learning method using K-Means clustering. The machine learning approach first represents each blog post as a binary vector indicating the presence of each word from the entire corpus. The K-Means algorithm then iteratively assigns each post to the nearest cluster center and recalculates centers as the average of assigned posts until assignments stabilize. The post demonstrates that roughly 40 lines of Ruby code can produce relevant recommendations and encourages readers to try libraries like Scikit Learn.
The Transformer architecture’s memory cost scales with token count, making long-context AI agents expensive. In controlled comparisons of three state space models - Gated KalmaNet, Gated DeltaNet, and Mamba-2 - Gated KalmaNet performed best, offering variable-length inference without retraining. Priming identifies which attention layers can be replaced by state space models using a sliding-window attention proxy, then reuses pre-trained weights to convert them into hybrid architectures without retraining, reducing pretraining cost by over 99%. A 32-billion-parameter hybrid model matches the original Transformer’s reasoning quality while decoding up to twice as fast or reaching correct answers in two-thirds of wall-clock time. The team extended hybrid context to 256K tokens without additional training and released primed models at 8B and 32B scales along with training code, kernels, and a vLLM plugin under Apache 2.0.
Cloudflare introduced Clef-omni, a decision model that natively handles audio, video, image, and text inputs in a single pipeline, built on a Qwen3-Omni mixture-of-experts backbone with text-to-speech components removed. Unlike cascading pipelines, it directly embeds media into a unified sequence and uses two-stage attention routing with lexical grammar to compute confidence scores, trained by freezing the backbone and applying LoRA adapters with label-smoothed cross-entropy and Brier score calibration. The company also cut Clef-flash’s price to $0.038 per million input tokens while reducing its context window from 64k to 24k for the hosted version, based on usage data showing only 0.24% of requests exceed that limit. Clef’s hosted version received speed optimizations at the infrastructure layer without new model weights, and Clef-omni is available open-weight on HuggingFace. Benchmarks show Clef-omni scoring 98.2 on BFCL case exact and 92.7 on API-Bank accuracy, with median text-only decision latency around 130 milliseconds.
Intercom launched a suite of new features including Custom Reports, Conversation Topics, and Macros to help teams balance personal and efficient customer communication at scale. Custom Reports lets teams tailor dashboards in just a few simple steps by building custom charts from user and conversation data attributes and exporting them. Conversation Topics provides a machine-learning powered engine that visually represents what customers are talking about and how often, giving teams an at-a-glance view of top concerns. Macros automate common inbox actions like tagging, assigning, snoozing, and closing conversations with a single click, boosting efficiency by up to 43%. Additional updates include Series targeting and split testing, enhanced Articles, mobile app tagging, permissions, and custom return paths for email deliverability.
Stripe makes thousands of near real-time risk decisions every second, balancing fraud detection against false positives that disrupt legitimate businesses. Zero-shot prompting achieved 0.487 precision, domain-guided prompting 0.706, and human reviewers 0.790 at the same recall. The team applied supervised fine-tuning (SFT) to an open-weight 4B-parameter LLM, serializing features like card payments and account activity into text records, training it on labels, and deriving a continuous fraud score from the next-token probabilities normalized over two label tokens. SFT increased average precision by approximately 24% over domain-guided prompting and 70% over zero-shot prompting, and consistently outperformed XGBoost across multiple risk tasks. Linear probing of hidden states showed that SFT made benign signals more linearly decodable, with ROC-AUC improving from 0.583 to 0.642, indicating the model better recognizes legitimate business activity and reduces false positives.
Existing systems estimate user price sensitivity from spending behavior or demographic proxies, missing residential property values and creating one-dimensional profiles. This invention proposes a four-step method that combines public property data with LLM-assisted data structuring, building-level price estimation, hierarchical clustering, and user mapping. The method collects public property records, uses LLMs to standardize attributes such as postcode, transaction price, and property size, then aggregates transactions within buildings to calculate weighted average sale prices. A hierarchical clustering model groups buildings into affluence tiers based on estimated resale values, accounting for regional market differences, and users are mapped to these tiers via aggregated geographic signals. The approach remains unimplemented and untested in live markets, with potential applications in personalization, advertising, and operational planning subject to future technical validation and privacy approvals.
Designed artifacts are ontological, shaping what becomes possible, yet systems enabling user authorship are typically evaluated on usability rather than ontological boundary negotiation. Two open-ended probes using a Wizard of Oz technique were created to let participants experience training a personalized machine learning system on phenomena they defined themselves. In a week-long exploratory study, participants used one of two probes during their everyday lives, allowing observation of boundary negotiations. Four sites of ontological boundary negotiation emerged: the boundaries of a phenomenon, the subject as part of relations, distinguishing signal from noise, and the objectivity of data. The paper offers starting points for supporting boundary negotiation through design and proposes open-ended probes as a method for ontological design.
Recent autonomous machine learning engineering agents increasingly rely on elaborate machinery such as multi-agent orchestrators and dedicated retrieval subagents, motivated by progress stagnation over long-horizon cycles and limited Large Language Model primitives. This paper finds that under an equal time budget and the same frontier LLM backbone, open-source state-of-the-art harnesses provide no advantages over a single session of a minimal-harness coding agent baseline that uses direct read, write, and bash primitives. The authors argue via large-scale systematic ablation studies that the machinery layers become redundant in the coding agent setting. The backbone, not the harness complexity, is the primary driver for performance. Consequently, the effort spent elaborating hand-crafted harnesses around strong models yields poor returns for current MLE benchmarks.
NVIDIA Kumo Tabular is an open foundation model for tabular classification and regression that predicts labels in a single forward pass without training, tuning, or feature engineering. It uses a Transformer architecture with column, row, and in-context attention, including length-aware attention temperature to maintain sharp attention on large tables. Pretrained entirely on artificial tables generated from structural causal models with realistic imperfections such as missing values, coarsened features, and heavy-tailed targets, it comes in three sizes (28M to 215M parameters) and handles missing values without imputation. On the TabArena, BeyondArena, TALENT, and ScoringBench benchmarks, Kumo Tabular achieves top rankings and establishes a new accuracy-efficiency frontier, running 17x faster than LimiX-2. The model is available on Hugging Face under the OpenMDW-1.1 license and runs via NVIDIA’s open-source structured-data-models library.
Human-in-the-loop (HITL) frameworks integrate human judgment into automated AI workflows to improve reliability and safety, but without careful design they can create bottlenecks. The post reveals key trade-offs: more automation increases speed and reduces cost but raises risk, while more human oversight boosts accuracy and safety but adds latency and cost. Common failure modes include over-reviewing trivial cases, lacking clear trigger logic, and expecting human reviewers to scale linearly. The proposed solution is a tiered HITL system applying the Pareto principle: automation handles 85% of high-confidence decisions using confidence calibration, with three tiers addressing the remaining cases under 3 ms latency, asynchronous review within hours, and real-time oversight with 30-60 second windows. Success requires careful routing, reviewer tooling, and closing feedback loops across seconds, days, weeks, and quarterly cycles to continuously improve model performance and reduce human involvement over time.
Accuracy is misleading for fraud detection because a model labeling everything legitimate can score 96% while catching zero fraud when fraud is only 4% of transactions. The post runs a hand-built Python grid search across algorithm, hyperparameters, feature set, class weight, and decision threshold, ranking by a custom cost metric balancing fraud savings against a US 6 analyst cost per flag. On a 480-transaction test split from a 2,400-transaction dataset with 4% fraud, the winning logistic regression caught 15 of 19 frauds, saving US 3,578 over a baseline that caught none. The winner’s evidence is passed to a local 1-billion-parameter LLM that drafts analyst notes, keeping the LLM downstream of the decision to avoid hallucination. The same workflow scales to Red Hat OpenShift AI by swapping local components for enterprise services like AutoML with AutoGluon and model serving via vLLM.
Apple’s on-device dictation tokenizer maps waveform to latent representations for a sparsely activated language model, and compressing it reduces memory and latency demands. The authors compress it via latent-space distillation, using the teacher’s pre-quantizer latent as the supervision target rather than tokens or output distributions. A student encoder is trained to regress the teacher’s per-frame latent with a squared-error objective, and an affine layer absorbs width mismatch. This recipe works for both token interfaces and for tokenizers pretrained alone or jointly with a language model, requiring no fine-tuning. At 2.8x compression, the student stays within 1.9% relative WER of its teacher on five of six pairs and outperforms an independently trained tokenizer of identical capacity by 3.9% relative.
BIRD is a research framework that converts five 2D perspective drawings into a proportionate, dimensioned 3D building envelope. It uses a dual-branch CNN to separate architectural line segments from noise, a text recognition model to scale them to real-world inches, and a wrapping algorithm to fold the elevation views around the floor plan. The framework achieved a median accuracy of roughly 93% across complete building plan sets from small residences to a multi-story apartment complex. The proposed deployment on MongoDB Atlas stores each project as a single JSON document containing all drawings’ segments, corrections, and final geometry, with S3 for image storage and Atlas Stream Processing for event-driven workflow coordination. MongoDB Search and Vector Search with Voyage AI embeddings enable full-text and semantic search over past projects, allowing drafters to start from corrected models of similar builds.
Amazon and Stanford University have launched the Stanford and Amazon Research Initiative, a formalized framework to advance research in AI, energy, and healthcare. The initiative builds on a deep relationship where currently more than 10 teams across Amazon fund active research and PhD fellowships at Stanford, spanning humanoid robotics, post-quantum cryptography, causal measurement science, and AI-driven radiology. It aims to accelerate the path from breakthrough research to real-world solutions that benefit society at large and broaden participation from diverse scholars. Focus areas will leverage strengths in artificial intelligence, machine learning, automated reasoning, and health, supported by Amazon’s cloud computing and AI services. The collaboration includes joint research projects, PhD fellowships, and interdisciplinary symposia, and was announced at a Stanford event with AWS CEO Matt Garman and Vice Provost David Studdert discussing the impact of university-industry collaborations.
Intelligent document processing (IDP) uses AI to extract, classify, validate, and structure information from PDFs, images, emails, forms, and other structured, semi-structured, or unstructured documents. Its pipeline combines document ingestion and preprocessing with OCR, document classification, NLP, layout analysis, data validation, human review for uncertain results, and delivery into downstream systems such as ERP, CRM, or accounting software. The post explains how embeddings support classification, how continuous learning adapts models to new layouts and corrections, and how metrics such as accuracy, throughput, confidence scores, and model drift support optimization. IDP can reduce manual effort, improve accuracy, lower costs, scale document-heavy workflows, and supply structured data for analytics, automation, AI agents, and retrieval-augmented generation, while high document variability and edge cases still make human-in-the-loop review relevant.
Tevogen Bio faced drug development costing upwards of $3 billion and requiring 10–12 years; selecting targets for its single-HLA-restricted SARS-COV2 product took 18–24 months of wet-lab work. Tevogen.AI partnered with Microsoft and Databricks to build and curate a multi-terabyte protein-sequence library and parallelize processing for machine-learning training. Databricks Data + AI Platform, Medallion Architecture, Unity Catalog, and distributed computing cut processing from 50 days to 24 hours under governed, fine-grained access controls. The dataset reached 24 million proteins, 16 billion datapoints, ~700 million unique peptides, and ~37 million cross-matched expert articles. An MLOps framework automated training, inference, monitoring, and retention; the alpha PredicTcell model used XGBoost and ESM models and reported 93–97% recall and 38–43% accuracy, while beta PredicTcell and alpha AdapTcell work began.
Activation steering has emerged as a way to guide generative models toward desired behaviors, but uniform interventions can reduce performance when steering is unnecessary. DSAS, or Dynamically Scaled Activation Steering, addresses this by separating when to steer from how to steer and adapting intervention strength across layers and inputs. At generation time, it computes context-dependent scaling factors that modify existing steering transformations, and it can also be jointly optimized end-to-end with the steering function. Combined with existing methods, DSAS improves the reported Pareto front between toxicity mitigation and utility preservation, while adding minimal computational overhead; experiments also apply it to a text-to-image diffusion model for modulating specific concepts and show token-level interpretability through the strength and location of steering.
Machine learning (ML) is a subset of artificial intelligence that learns statistical patterns from data to make predictions or decisions without hard-coded rules. The guide distinguishes ML from generative AI, which creates new content, and from deep learning, a specialized ML approach using multilayered neural networks that learns features automatically and typically requires more data and computing power. It describes a workflow of preparing and splitting data, training models by comparing predictions with actual outcomes, and adjusting parameters to reduce error, illustrated with image classification and customer-churn scoring. It also covers supervised, unsupervised, and reinforcement learning, applications including computer vision and natural language processing, newer architectures such as transformers and Mamba, and MLOps practices for deployment, monitoring, and retraining.
The best engineering reads, once a week.
