Transformer-based Large Language Models (LLMs) have become increasingly important. However, due to the quadratic time complexity of attention computation, scaling LLMs to longer contexts incurs extremely slow inference latency and high GPU memory consumption for caching key-value (KV) vectors. This paper proposes RetrievalAttention, a training-free approach to both accelerate attention computation and reduce GPU memory consumption. By leveraging the dynamic sparsity of attention mechanism, RetrievalAttention proposes to use approximate nearest neighbor search (ANNS) indexes for KV vectors in CPU memory and retrieves the most relevant ones with vector search during generation. Unfortunately, we observe that the off-the-shelf ANNS indexes are often ineffective for such retrieval tasks due to the out-of-distribution (OOD) between query vectors and key vectors in attention mechanism. RetrievalAttention addresses the OOD challenge by designing an attention-aware vector search algorithm that can adapt to the distribution of query vectors. Our evaluation shows that RetrievalAttention only needs to access 1--3% of data while maintaining high model accuracy. This leads to significant reduction in the inference cost of long-context LLMs with much lower GPU memory footprint. In particular, RetrievalAttention only needs a single NVIDIA RTX4090 (24GB) for serving 128K tokens in LLMs with 8B parameters, which is capable of generating one token in 0.188 seconds.

There has been much recent interest in designing symmetry-aware neural networks (NNs) exhibiting relaxed equivariance. Such NNs aim to interpolate between being exactly equivariant and being fully flexible, affording consistent performance benefits. In a separate line of work, certain structured parameter matrices -- those with displacement structure, characterized by low displacement rank (LDR) -- have been used to design small-footprint NNs. Displacement structure enables fast function and gradient evaluation, but permits accurate approximations via compression primarily to classical convolutional neural networks (CNNs). In this work, we propose a general framework -- based on a novel construction of symmetry-based structured matrices -- to build approximately equivariant NNs with significantly reduced parameter counts. Our framework integrates the two aforementioned lines of work via the use of so-called Group Matrices (GMs), a forgotten precursor to the modern notion of regular representations of finite groups. GMs allow the design of structured matrices -- resembling LDR matrices -- which generalize the linear operations of a classical CNN from cyclic groups to general finite groups and their homogeneous spaces. We show that GMs can be employed to extend all the elementary operations of CNNs to general discrete groups. Further, the theory of structured matrices based on GMs provides a generalization of LDR theory focussed on matrices with cyclic structure, providing a tool for implementing approximate equivariance for discrete groups. We test GM-based architectures on a variety of tasks in the presence of relaxed symmetry. We report that our framework consistently performs competitively compared to approximately equivariant NNs, and other structured matrix-based compression frameworks, sometimes with a one or two orders of magnitude lower parameter count.

Despite significant progress in static protein structure collection and prediction, the dynamic behavior of proteins, one of their most vital characteristics, has been largely overlooked in prior research. This oversight can be attributed to the limited availability, diversity, and heterogeneity of dynamic protein datasets. To address this gap, we propose to enhance existing prestigious static 3D protein structural databases, such as the Protein Data Bank (PDB), by integrating dynamic data and additional physical properties. Specifically, we introduce a large-scale dataset, Dynamic PDB, encompassing approximately 12.6K proteins, each subjected to all-atom molecular dynamics (MD) simulations lasting 1 microsecond to capture conformational changes. Furthermore, we provide a comprehensive suite of physical properties, including atomic velocities and forces, potential and kinetic energies of proteins, and the temperature of the simulation environment, recorded at 1 picosecond intervals throughout the simulations. For benchmarking purposes, we evaluate state-of-the-art methods on the proposed dataset for the task of trajectory prediction. To demonstrate the value of integrating richer physical properties in the study of protein dynamics and related model design, we base our approach on the SE(3) diffusion model and incorporate these physical properties into the trajectory prediction process. Preliminary results indicate that this straightforward extension of the SE(3) model yields improved accuracy, as measured by MAE and RMSD, when the proposed physical properties are taken into consideration. https://fudan-generative-vision.github.io/dynamicPDB/ .

Relational representation learning transforms relational data into continuous and low-dimensional vector representations. However, vector-based representations fall short in capturing crucial properties of relational data that are complex and symbolic. We propose geometric relational embeddings, a paradigm of relational embeddings that respect the underlying symbolic structures. Specifically, this dissertation introduces various geometric relational embedding models capable of capturing: 1) complex structured patterns like hierarchies and cycles in networks and knowledge graphs; 2) logical structures in ontologies and logical constraints applicable for constraining machine learning model outputs; and 3) high-order structures between entities and relations. Our results obtained from benchmark and real-world datasets demonstrate the efficacy of geometric relational embeddings in adeptly capturing these discrete, symbolic, and structured properties inherent in relational data.

Two people have been arrested on suspicion of murder after a baby died in Stoke-on-Trent. Police and the ambulance service were called to Sherwin Road, Burslem, shortly after 09:00 BST on 27 August following the death of a baby boy. A woman, 26, and man, 25, were arrested on 30 August on suspicion of causing or allowing the death of a child. The pair, from Stoke-on-Trent, were further arrested on suspicion of murder on Tuesday, Staffordshire Police said. A spokesperson for the force said specialist officers were supporting the baby's family.

Inventor and futurist Ray Kurzweil, researcher and Brookings Institution fellow Chinasa T. Okolo, director of the U.S. Artificial Safety Institute (AISI) Elizabeth Kelly, and Cognizant CEO Ravi Kumar S, discussed the transformative power of AI during a panel at a TIME100 Impact Dinner in San Francisco on Monday. During the discussion, which was moderated by TIME's editor-in-chief Sam Jacobs, Kurzweil predicted that we will achieve Artificial General Intelligence (AGI), a type of AI that might be smarter than humans, by 2029. "Nobody really took it seriously until now," Kurzweil said about AI. "People are convinced it's going to either endow us with things we'd never had before, or it's going to kill us." Cognizant sponsored Monday's event, which celebrated the 100 most influential people leading change in AI. Jacobs probed the four panelists--three of whom were named to the 2024 list--about the opportunities and challenges presented by AI's rapid advancement.

Quantum computers represent a transformative frontier in computational technology, promising exponential speedups beyond classical computing limits. IBM Quantum has led significant advancements in both hardware and software, providing access to quantum hardware via IBM Cloud since 2016, achieving a milestone with the world's first accessible quantum computer. This article explores IBM's quantum computing journey, focusing on the development of practical quantum computers. We summarize the evolution and advancements of IBM Quantum's processors across generations, including their recent breakthrough surpassing the 1,000-qubit barrier. The paper reviews detailed performance metrics across various hardware, tracing their evolution over time and highlighting IBM Quantum's transition from the noisy intermediate-scale quantum (NISQ) computing era towards fault-tolerant quantum computing capabilities.

Graph neural networks (GNNs) are a type of neural network capable of learning on graph-structured data. However, training GNNs on large-scale graphs is challenging due to iterative aggregations of high-dimensional features from neighboring vertices within sparse graph structures combined with neural network operations. The sparsity of graphs frequently results in suboptimal memory access patterns and longer training time. Graph reordering is an optimization strategy aiming to improve the graph data layout. It has shown to be effective to speed up graph analytics workloads, but its effect on the performance of GNN training has not been investigated yet. The generalization of reordering to GNN performance is nontrivial, as multiple aspects must be considered: GNN hyper-parameters such as the number of layers, the number of hidden dimensions, and the feature size used in the GNN model, neural network operations, large intermediate vertex states, and GPU acceleration. In our work, we close this gap by performing an empirical evaluation of 12 reordering strategies in two state-of-the-art GNN systems, PyTorch Geometric and Deep Graph Library. Our results show that graph reordering is effective in reducing training time for CPU- and GPU-based training, respectively. Further, we find that GNN hyper-parameters influence the effectiveness of reordering, that reordering metrics play an important role in selecting a reordering strategy, that lightweight reordering performs better for GPU-based than for CPU-based training, and that invested reordering time can in many cases be amortized.

Next-token prediction serves as the dominant component in current neural language models. During the training phase, the model employs teacher forcing, which predicts tokens based on all preceding ground truth tokens. However, this approach has been found to create shortcuts, utilizing the revealed prefix to spuriously fit future tokens, potentially compromising the accuracy of the next-token predictor. In this paper, we introduce Semformer, a novel method of training a Transformer language model that explicitly models the semantic planning of response. Specifically, we incorporate a sequence of planning tokens into the prefix, guiding the planning token representations to predict the latent semantic representations of the response, which are induced by an autoencoder. In a minimal planning task (i.e., graph path-finding), our model exhibits near-perfect performance and effectively mitigates shortcut learning, a feat that standard training methods and baseline models have been unable to accomplish. Furthermore, we pretrain Semformer from scratch with 125M parameters, demonstrating its efficacy through measures of perplexity, in-context learning, and fine-tuning on summarization tasks.

The modern theory of classical mechanics, developed by Lagrange, Hamilton and Noether, attempts to cast all of classical motion in the form of an optimization problem, based on an energy functional called the classical action. The most important advantage of this formalism is the ability to manifestly incorporate and exploit symmetries and conservation laws. This reformulation succeeded for unconstrained and holonomic systems that at most obey position equality constraints. Non-holonomic systems, which obey velocity dependent constraints or position inequality constraints, are abundant in nature and of central relevance for science, engineering and industry. All attempts so far to solve non-holonomic dynamics as a classical action optimization problem have failed. Here we utilize the classical limit of a quantum field theory action principle to construct a novel classical action for non-holonomic systems. We therefore put to rest the 190 year old question of whether classical mechanics is variational, answering in the affirmative. We illustrate and validate our approach by solving three canonical model problems by direct numerical optimization of our new action. The formalism developed in this work significantly extends the reach of action principles to a large class of relevant mechanical systems, opening new avenues for their analysis and control both analytically and numerically.