Depth Pro 2: Sharp and Temporally Consistent Metric Monocular DepthWe present Depth Pro 2, an approach to estimating dense metric depth for images and video at full resolution. By analyzing prior methods for monocular depth estimation, we find that they commonly trade off temporal consistency and boundary Apple ML11/20 08:00
Split-and-Scale Latent 3D RepresentationsLatent 3D representations have significantly expanded the capabilities of modern shape modeling, enabling compact encoding of 3D scenes and supporting powerful 3D generative models. Despite this progress, current representations struggle toApple ML11/16 08:00
Theoretical Limits of Language Model AlignmentLanguage model (LM) alignment improves model outputs to reflect human preferences while preserving the capabilities of the base model. The most common alignment approaches are (i) reinforcement learning, which maximizes the expected reward Apple ML11/12 08:00
Evaluating the Evaluators: Investigating LLM Judges for Personalized Writing Style AssessmentThis paper was accepted at the Can We Trust AI Evaluation? Workshop at NeurIPS 2026. We evaluate how good large language model (LLM) judges are for personalized writing tasks. This is important when a user’s preferred writing style is knownApple ML11/10 08:00
Understanding Private Evolution as Learning-Augmented ClusteringPrivate Evolution (PE) is a differentially private algorithm for synthetic data generation. While it can be viewed as a Wasserstein learning algorithm, it performs much better in practice than worst-case Wasserstein analyses would predict. Apple ML11/10 08:00
On the Importance of Gating: Memorization vs. In-Context Learning in State Space ModelsState Space Models (SSMs) have emerged as a compelling alternative to Transformers, enabling sequence modeling with constant memory and linear compute. Although SSMs exhibit reasonable performance and favorable computational characteristicsApple ML11/10 08:00
Weighted State Souping for Efficient Long-Context RetrievalThis paper was accepted at the Workshop on Personalized, Aligned, Long-term Memory for AI Systems (PALM) at NeurIPS 2026. State space models (SSMs) combine fast inference and fixed state size with competitive language-modeling performance. Apple ML11/10 08:00
Toward a Theory of EvaluabilityMachine learning models, particularly large foundation models, are increasingly evaluated along multiple axes that are relevant for their usage. In addition to accuracy measures appropriate for the specific task of interest, measures capturApple ML11/10 08:00
Complexity of Differentially Private Selection via Federated APIsDifferential privacy (DP) has become the standard for privacy-preserving data analysis, yet its practical implementation in federated or distributed settings often faces a significant utility gap compared to the central model. While existinApple ML11/10 08:00
Faster Retrieval for Better, More Efficient RAGMaximum inner product search (MIPS) sits at the core of recommendation systems, search, and retrieval-augmented generation (RAG). Given a query vector, MIPS systems search within a database of millions of candidate key vectors for the one tApple ML11/05 08:00
Attention to Mamba: A Recipe for Cross-Architecture DistillationState Space Models (SSMs) such as Mamba have become a popular alternative to Transformer models, due to their reduced memory consumption and higher throughput at generation compared to their Attention-based counterparts. On the other hand, Apple ML11/03 08:00
GHARP: Real-Time Gaussian Head Animation from Large-Scale Reconstruction PriorWe present GHARP (Real-time Gaussian Head Animation from Large-scale Reconstruction Prior), a method that animates 3D human heads in real time from a few input images of a subject and a driving expression signal. We decouple the problem intApple ML10/26 08:00
Pixels to Tokens: Token Space Efficient Active Learning for Low-Budget Semantic SegmentationWhile foundation models and vision transformers (ViTs) have redefined visual representations, current active learning (AL) strategies for these backbones largely operate at the patch or region level, where annotation requirements remain higApple ML10/21 08:00
Plan or Retrieve First? A Scale Study of Tool Discovery ArchitecturesTool-augmented LLM agents discover relevant APIs from large registries before planning execution, yet finite context windows prohibit injecting all tool definitions into a single prompt. The dominant approach, which we term Retrieve-then-PlApple ML10/15 08:00
Selection-Based Structured Reasoning: Toward Efficient Multimodal Search AgentsMultimodal agents commonly generate freeform reasoning before each action. For small models, limited model capacity can result in lengthy reasoning that provides little useful guidance for action generation while incurring substantial inferApple ML10/15 08:00
Local and Distributed Large LLM Inference with MLX and RDMA on Mac StudioMLX on Mac with Apple silicon lets AI researchers and developers experiment locally with the latest models and methods in a private environment. The open-source array framework leverages the unified memory architecture of Apple silicon, enaApple ML10/15 08:00
The Design Space of Tri-Modal Masked Diffusion ModelsDiscrete diffusion models have emerged as strong alternatives to autoregressive language models, with recent work initializing and finetuning a base unimodal model for bi-modal generation. Diverging from previous approaches, we introduce thApple ML10/14 08:00
Code-Switching Detection via Language-Selective Dimensions in Speech Foundation ModelsCode-switching (CS), the alternation between languages within an utterance, is common in multilingual speech but underrepresented in pretraining corpora. Increasing CS coverage requires finding relevant utterances in large speech collectionApple ML10/14 08:00
Mix, Don’t Tune: Bilingual Pre-Training Outperforms Hyperparameter Search in Data-Constrained SettingsFor most languages of the world, language model pre-training operates in a data-constrained regime where models must repeat their training data many times, degrading generalization. Two remedies exist: aggressive hyperparameter tuning such Apple ML10/13 08:00
Large Language Models for Machine Translation Quality Annotation: Humans and Models Are Both ChallengedLarge Language Models (LLMs) are considered to be a more efficient and cost-effective alternative to human judgment for Machine Translation (MT) evaluation. With MT evaluation spanning a large number of language pairs, domains and levels ofApple ML10/12 08:00