Tag
This paper formulates a multi-item capacitated lot-sizing problem with stochastic demand timing as a discrete-time MDP and proposes a genetic algorithm to solve it, demonstrating efficiency on benchmark instances.
This paper identifies aggregation-induced reward hacking in multi-reward reinforcement learning for LLMs and proposes an adaptive projection method, AMRP, to dynamically adjust weights for better reward balance and performance.
The paper investigates percolation dynamics in optimization, with a focus on variance cascades and discrete scale invariance.
The article explains the efficient frontier concept in LLM inference, covering tradeoffs between latency, throughput, and cost, and outlines techniques to manage or enhance efficiency in deployment.
Researchers from Washington State University used AI to optimize 3D printing of the high-performance metal alloy GRCop-42, making it accessible with commercial equipment. The work was published in AAAI proceedings and won an Innovative Deployed Application Award.
The PyTorchCon North America 2026 features a Kernel Engineering track focused on compilers, optimization, custom kernels, and systems work, with a keynote by Mark Saroufim.
The article discusses the challenges of achieving low latency, specifically a 200-millisecond threshold, in enterprise AI applications, highlighting its impact on real-time performance and operational efficiency.
The article evaluates whether minifying CSS is necessary for web performance, concluding that the benefits are negligible based on tests and practical analysis.
The article discusses hidden sources of latency in voice-agent systems after streaming is enabled and seeks advice on which performance metrics are crucial for real-time agents.
An article explaining static, dynamic, and continuous batching strategies in LLM inference, and why serving LLMs differs from traditional ML inference.
SS-eSOAP introduces a self-scaled adaptive preconditioning method for physics-informed neural networks, improving optimization by reducing residuals and achieving higher accuracy on various PDE benchmarks compared to baseline methods.
This paper proposes a fully distributed continuous-time algorithm for computing Generalized Nash Equilibria in multi-agent systems with shared constraints, eliminating the need for multiplier exchange to reduce communication overhead and enhance privacy, validated on a multi-robot placement task.
This thesis elucidates mode connectivity in neural network loss landscapes through spectral dynamics, explaining phenomena like grokking and memorization, and applies insights to enhance diverse feature learning and practical neural network training.
The paper proposes TREVIS, a method that uses a Tree Transformer Variational Auto-Encoder to learn sparse decision trees by optimizing in a continuous latent space, achieving good predictive performance with improved structural sparsity.
A GitHub repository providing a structured 100-day learning plan for LLM inference engineering, covering topics from CUDA kernels to autoscaling, with runnable scripts.
David Cortés is building a site to showcase workflows using pi-autoresearch and is inviting optimization runs to embed on the site.
Demonstrates that Qwen 3.8 Flash Next can run locally on a mid-range Android phone at 3.5 tokens per second with optimizations and low quantization.
The tweet advises using different AI models for tasks of varying complexity to optimize results, suggesting Sol for complex work, Terra for daily tasks, and Luna for simple tasks.
A tweet asking about the speed of an unspecified system, with the author sharing their personal best performance of 1.6 meters per second.
A call for contributions to open pull requests in llama.cpp aimed at optimizing CPU, RAM, disk, and hybrid inference for faster performance, targeting completion by year-end.