Apple at ICLR 2026 in Rio: over 40 posters, MLX demo on iPad Pro, SHARP 3D generation and MANZANO unified model

Apple Machine Learning Research published on April 17, 2026 an overview of research the company is presenting at ICLR 2026 in Rio de Janeiro. The conference runs from April 23 to 27, and Apple is not only a sponsor but also institutionally engaged through area chairs and workshop organisers.

What Apple is presenting

Apple organises its announcement across main research domains:

Computer vision

• “Less Gaussians, Texture More” — advances in 3D Gaussian splatting technique for scene reconstruction from images
• SHARP — monocular view synthesis (creates new viewing angles from a single image)
• MANZANO — unified multimodal model integrating visual understanding, generation and reasoning

Language models

• Hallucination detection — how to measure and prevent AI fabrications
• Semantic calibration of LLMs — how models assess their own accuracy
• Reasoning enhancement through reinforcement learning

AI safety

• Joint multimodal safety assessment — evaluating safety of models that combine text, image and audio
• Interpretability frameworks — understanding the internal representations of models

Efficiency

• Quantization-aware training — reducing models without quality loss
• Parallel RNN processing — alternatives to the Transformer paradigm

Scale of presence

Apple has a respectable presence:

• More than 40 poster sessions
• One oral presentation: “To Infinity and Beyond — Tool-Use Unlocks Length Generalization”
• Workshop papers across all five days of the conference

Oral presentations at ICLR are highly competitive — roughly 2–3 percent of submissions are accepted — so one Apple oral presentation is significant.

Booth 204 — live demonstrations

Apple has traditionally taken a restrained approach to public ML demonstrations, but at ICLR 2026 that changes. At booth 204 they show:

• Local LLM inference on Apple silicon through the MLX framework — Apple’s performant ML framework open-sourced last year
• SHARP 3D generation live on iPad Pro — a real-time demonstration of a demanding model on a mobile device

This is a strategic signal — Apple wants ML researchers to see how Apple silicon can execute production workloads.

Leadership engagement

Apple employees hold significant conference roles:

• Area chairs (selecting which papers are accepted)
• Workshop organisers

This reflects Apple’s long-term institutional connection with the ML research community — it is not just a PR presence, but substantive academic engagement.

Broader context

Apple has historically been less visible in ML publications than Google or Meta — the company has traditionally favoured internal research that manifests through products, rather than public papers. That has been changing over the past few years with the MLX framework, regular Apple ML Research blog posts, and participation in major conferences.

ICLR 2026 in Rio represents one of Apple’s largest ML presences to date. For researchers tracking the Apple trend, 40+ posters + one oral presentation + live MLX demos sends a clear message: Apple is positioning itself as a serious academic partner, not just a product company that occasionally releases a paper.

For practitioners considering the MLX framework and Apple silicon for ML workloads, the conference will be interesting to follow — especially if you have the opportunity to see the live demo at the booth.