Vahab Mirrokni

VP, Google Fellow@Google Research·美国·OpenReview

研究方向

Economics and Computation · Graph-based Learning · ML Optimization · GenAI Data · ML artichecture and LLM Modeling · Data Mining and Clustering at Scale

Vahab Mirrokni

TNT: Improving Chunkwise Training for Test-Time Memorization

Understanding the Role of Training Data in Test-Time Scaling

TurboQuant: Online Vector Quantization with Near-optimal Distortion Rate

ATLAS: Learning to Optimally Memorize the Context at Test Time

Lattice: Learning to Efficiently Compress the Memory

Data Selection for Fine-tuning Vision Language Models via Cross Modal Alignment Trajectories

Memory Caching: RNNs with Growing Memory

It's All Connected: A Journey Through Test-Time Memorization, Attentional Bias, Retention, and Online Optimization

Shared Embedding Optimization: A Two-Stage Approach for Efficient and Effective Feature Embedding

Language Models Need Sleep: Learning to Self Modify and Consolidate Memories

Data Selection and Active Learning via Low-Rank Approximation

Titans: Learning to Memorize at Test Time

Self-Boost via Optimal Retraining: An Analysis via Approximate Message Passing

PiKE: Adaptive Data Mixing for Large-Scale Multi-Task Learning Under Low Gradient Conflicts

TRELLIS: Learning to Compress Key-Value Memory in Attention Models

Procurement Auctions via Approximately Optimal Submodular Optimization

DiSK: Differentially Private Optimizer with Simplified Kalman Filter for Noise Reduction

Retraining with Predicted Hard Labels Provably Increases Model Accuracy

MS-SSM: A Multi-Scale State Space Model for Efficient Sequence Modeling

Maximum Coverage in Turnstile Streams with Applications to Fingerprinting Measures

Nested Learning: The Illusion of Deep Learning Architectures

Efficient Data Selection at Scale via Influence Distillation

ClusterGen: Token Generation in Sublinear Time and Memory with Clustering KV Cache

Best of Both Worlds: Advantages of Hybrid Graph Sequence Models

Improving the Variance of Differentially Private Randomized Experiments through Clustering

Addax: Utilizing Zeroth-Order Gradients to Improve Memory Efficiency and Performance of SGD for Fine-Tuning Language Models

DeepCrossAttention: Supercharging Transformer Residual Connections

Maximum Coverage in Turnstile Streams with Applications to Fingerprinting Measures

Synthetic Text Generation for Training Large Language Models via Gradient Matching

Towards an Understanding of Graph Sequence Models

NanoMoE: Scaling Mixture of Experts to Individual Layers for Parameter-Efficient Deep Learning

Retraining with Predicted Hard Labels Provably Increases Model Accuracy