AI & Machine Learning

Provides a robust solution for anti-aliasing in Feed-forward Gaussian Splatting, enabling high-fidelity rendering across varying sampling rates and resolutions.

Uses token-level perplexity analysis to prove that LLMs rely on simple heuristics rather than the linguistic reasoning they appear to exhibit on standard benchmarks.

Demonstrates that most 'failures' of AI agents on data engineering benchmarks are actually due to flawed ground-truth and rigid evaluation scripts rather than model inability.

Enables precise Camera-LiDAR extrinsic calibration even under massive initial misalignments that typically break automated calibration systems.

Shows that VLMs can overcome deep-seated perceptual biases and optical illusions by using image manipulation tools rather than more training data.

Obtain epistemic and aleatoric uncertainty from a single forward-backward pass of an unmodified pretrained LLM.

The first prior-fitted foundation model for survival analysis that enables zero-shot time-to-event predictions on tabular data.

Mathematical proof that cosine similarity between label representations (unembeddings) in softmax classifiers is fundamentally uninformative.

A vector-wise sparse attention mechanism that accelerates long-context video inference by 2.6x with zero loss in accuracy.

A novel neural primitive based on metriplectic dynamics that outperforms Transformers in data efficiency and generalization.

A debunking of the idea that single-vector embedding failures are primarily due to low dimensionality.

A unified quantization and runtime framework for deploying multiple LoRA-adapted generative models on edge devices simultaneously.

A diagnostic revealing that over 50% of video understanding benchmark samples can be solved without any video or temporal context.

A 1D continuous image tokenizer that uses semantic masking to achieve a 64x reduction in token usage without sacrificing generation fidelity.

A unified agentic framework that closes the 'AI-for-AI' research loop by discovering novel architectures, data pipelines, and algorithms.

Introduces the 'near-miss' metric to detect latent failures in agentic workflows where agents bypass policy checks but reach correct outcomes by chance.

A compiler approach to agent logs that reduces token consumption by 50-66% while improving context learning performance.

Provides a closed-form safety law for Dynamic Movement Primitives, enabling provably safe robot control without real-time optimization.

A training-free attack that removes diffusion-based watermarks with nearly 100% success by deflecting the generative trajectory.

A stabilization mechanism for adapting LLMs to time-series tasks that reduces memory footprint by up to 1,776x.

Identifies a 'dual-capability bottleneck' where low-rated training data is essential for state tracking while high-rated data is needed for decision quality.

A novel approach to upcycle multiple dense expert models into a unified Mixture-of-Experts model without any additional training.

Provides a computationally efficient 'early warning' system for emergent capabilities like grokking and induction head formation using 2-datapoint reduced density matrices.

Introduces a GUI-native agent system that operates complex scientific instruments through their existing visual interfaces rather than requiring proprietary APIs.

Decouples high-level intent planning from low-level motor control in Vision-Language-Action (VLA) models to prevent the degradation of pre-trained VLM representations.

Demonstrates that independent aggregation (Hybrid Confirmation Tree) consistently outperforms the standard 'AI-as-advisor' paradigm across diverse high-stakes domains.

Applies Shapley values from cooperative game theory to solve the 'free-rider' problem in GRPO-based reinforcement learning post-training.

Proves that complex GraphRAG systems can be simplified into a more efficient 'UnWeaver' framework that achieves the same benefits using entity-based decomposition and standard VectorRAG.

Identifies 'label leakage' from limited task diversity as the primary bottleneck for relational foundation models, rather than raw data volume.

Shows that deep learning models for medical imaging (MRI) can be trained using synthetic quaternion Julia fractals instead of sensitive human clinical data.

Produces high-fidelity SHAP explanations for tabular data 1000x faster than traditional methods by integrating them directly into the model architecture.

Provides a formal framework for optimizing models whose decisions actively change the distribution of the data they encounter.

Introduces a rigorous algorithm to determine if two different neural networks share the same underlying 'algorithmic interpretation' without needing to manually define the circuits.

Replaces heuristic ReAct-style agent loops with a mathematical framework based on control theory to prevent LLM agents from over-deliberating or using excessive tools.

Proposes a unified tensor-factorization view of attention that encompasses MHA, GQA, and MLA while reducing parameter counts by an order of magnitude.

Identifies the specific conditions under which Reinforcement Learning causes LLMs to 'lie' or hide reasoning in their Chain-of-Thought (CoT).

Discovers that video diffusion models commit to high-level plans in the first few denoising steps, enabling a new inference-time scaling technique called ChEaP.

Researchers have built computer chips that can run 'backward' to solve math problems that are normally impossible for modern hardware.

A famous musical masterpiece was found to be so mathematically perfect that an algorithm can reconstruct 93% of the score from scratch.

Scientists are using the high-level math of 'cohomology'—usually used to describe the shape of the universe—to find bugs in computer code.

Researchers can now predict the exact moment an AI agent will go 'rogue' and leak data before it actually happens.

AI has learned to objectively measure the 'groove' and funkiness of music, outperforming traditional human-designed formulas.

A new cyberattack can physically destroy an AI chip simply by changing the order in which it adds numbers.

AI coding agents are actually safer than human programmers at building new software, but twice as dangerous when it comes to maintaining it.

Researchers believe they have discovered a new transcendental number as fundamental as Pi or e.

Sanctioned crypto users are evading asset freezes by 'bribing' the computers that process blockchain transactions.

A standard graphics card can now track the entire Starlink satellite network in less than four milliseconds.

A new computer chip uses the quantum flipping of individual electron spins to solve physics problems that are too 'random' for standard processors.

Researchers have mapped out the physical limits of quantum teleportation, revealing how many particles can be 'beamed' before the signal collapses.

Scientists have developed a way to control swarms of microscopic nanorobots inside the body without needing to talk to them individually.