Mixture of Hidden-Dimensions Transformer

Transformer models encounter challenges in scaling hidden dimensions efficiently, as uniformly increasing them inflates computational and memory costs while failing to emphasize the most relevant features for each token. For further understanding, we study hidden dimension sparsity and observe that trained Transformers utilize only a small fraction of token dimensions, revealing an "activation flow" pattern. Notably, there are shared sub-dimensions with sustained activation across multiple consecutive tokens and specialized sub-dimensions uniquely activated for each token. To better model token-relevant sub-dimensions, we propose MoHD (Mixture of Hidden Dimensions), a sparse conditional activation architecture. Particularly, MoHD employs shared sub-dimensions for common token features and a routing mechanism to dynamically activate specialized sub-dimensions. To mitigate potential information loss from sparsity, we design activation scaling and group fusion mechanisms to preserve activation flow. In this way, MoHD expands hidden dimensions with negligible increases in computation or parameters, efficient training and inference while maintaining performance. Evaluations across 10 NLP tasks show that MoHD surpasses Vanilla Transformers in parameter efficiency and task performance. It achieves 1.7% higher performance with 50% fewer activation parameters and 3.7% higher performance with a 3x parameter expansion at constant activation cost. MOHD offers a new perspective for scaling the model, showcasing the potential of hidden dimension sparsity to boost efficiency

Harnessing Your DRAM and SSD for Sustainable and Accessible LLM Inference with Mixed-Precision and Multi-level Caching

Although Large Language Models (LLMs) have demonstrated remarkable capabilities, their massive parameter counts and associated extensive computing make LLMs' deployment the main part of carbon emission from nowadays AI applications. Compared to modern GPUs like H$100$, it would be significantly carbon-sustainable if we could leverage old-fashioned GPUs such as M$40$ (as shown in Figure 1, M$40$ only has one third carbon emission of H$100$'s) for LLM servings. However, the limited High Bandwidth Memory (HBM) available on such GPU often cannot support the loading of LLMs due to the gigantic model size and intermediate activation data, making their serving challenging. For instance, a LLaMA2 model with $70$B parameters typically requires $128$GB for inference, which substantially surpasses $24$GB HBM in a $3090$ GPU and remains infeasible even considering the additional $64$GB DRAM. To address this challenge, this paper proposes a mixed-precision with a model modularization algorithm to enable LLM inference on outdated hardware with resource constraints. (The precision denotes the numerical precision like FP16, INT8, INT4) and multi-level caching (M2Cache).) Specifically, our M2Cache first modulizes neurons in LLM and creates their importance ranking. Then, it adopts a dynamic sparse mixed-precision quantization mechanism in weight space to reduce computational demands and communication overhead at each decoding step. It collectively lowers the operational carbon emissions associated with LLM inference. Moreover, M2Cache introduces a three-level cache management system with HBM, DRAM, and SSDs that complements the dynamic sparse mixed-precision inference. To enhance communication efficiency, M2Cache maintains a neuron-level mixed-precision LRU cache in HBM, a larger layer-aware cache in DRAM, and a full model in SSD.

A Cluster-Based Statistical Channel Model for Integrated Sensing and Communication Channels

The emerging 6G network envisions integrated sensing and communication (ISAC) as a promising solution to meet growing demand for native perception ability. To optimize and evaluate ISAC systems and techniques, it is crucial to have an accurate and realistic wireless channel model. However, some important features of ISAC channels have not been well characterized, for example, most existing ISAC channel models consider communication channels and sensing channels independently, whereas ignoring correlation under the consistent environment. Moreover, sensing channels have not been well modeled in the existing standard-level channel models. Therefore, in order to better model ISAC channel, a cluster-based statistical channel model is proposed in this paper, which is based on measurements conducted at 28 GHz. In the proposed model, a new framework based on 3GPP standard is proposed, which includes communication clusters and sensing clusters. Clustering and tracking algorithms are used to extract and analyze ISAC channel characteristics. Furthermore, some special sensing cluster structures such as shared sensing cluster, newborn sensing cluster, etc., are defined to model correlation and difference between communication and sensing channels. Finally, accuracy of the proposed model is validated based on measurements and simulations.

Non-stationarity Characteristics in Dynamic Vehicular ISAC Channels at 28 GHz

Integrated sensing and communications (ISAC) is a potential technology of 6G, aiming to enable end-to-end information processing ability and native perception capability for future communication systems. As an important part of the ISAC application scenarios, ISAC aided vehicle-to-everything (V2X) can improve the traffic efficiency and safety through intercommunication and synchronous perception. It is necessary to carry out measurement, characterization, and modeling for vehicular ISAC channels as the basic theoretical support for system design. In this paper, dynamic vehicular ISAC channel measurements at 28 GHz are carried out and provide data for the characterization of non-stationarity characteristics. Based on the actual measurements, this paper analyzes the time-varying PDPs, RMSDS and non-stationarity characteristics of front, lower front, left and right perception directions in a complicated V2X scenarios. The research in this paper can enrich the investigation of vehicular ISAC channels and enable the analysis and design of vehicular ISAC systems.

Channel Measurements and Modeling for Dynamic Vehicular ISAC Scenarios at 28 GHz

Integrated sensing and communication (ISAC) is a promising technology for 6G, with the goal of providing end-to-end information processing and inherent perception capabilities for future communication systems. Within ISAC emerging application scenarios, vehicular ISAC technologies have the potential to enhance traffic efficiency and safety through integration of communication and synchronized perception abilities. To establish a foundational theoretical support for vehicular ISAC system design and standardization, it is necessary to conduct channel measurements, and modeling to obtain a deep understanding of the radio propagation. In this paper, a dynamic statistical channel model is proposed for vehicular ISAC scenarios, incorporating Sensing Multipath Components (S-MPCs) and Clutter Multipath Components (C-MPCs), which are identified by the proposed tracking algorithm. Based on actual vehicular ISAC channel measurements at 28 GHz, time-varying sensing characteristics in front, left, and right directions are investigated. To model the dynamic evolution process of channel, number of new S-MPCs, lifetimes, initial power and delay positions, dynamic variations within their lifetimes, clustering, power decay, and fading of C-MPCs are statistically characterized. Finally, the paper provides implementation of dynamic vehicular ISAC model and validates it by comparing key simulation statistics between measurements and simulations.

Characterization of Wireless Channel Semantics: A New Paradigm

Recently, deep learning enabled semantic communications have been developed to understand transmission content from semantic level, which realize effective and accurate information transfer. Aiming to the vision of sixth generation (6G) networks, wireless devices are expected to have native perception and intelligent capabilities, which associate wireless channel with surrounding environments from physical propagation dimension to semantic information dimension. Inspired by these, we aim to provide a new paradigm on wireless channel from semantic level. A channel semantic model and its characterization framework are proposed in this paper. Specifically, a channel semantic model composes of status semantics, behavior semantics and event semantics. Based on actual channel measurement at 28 GHz, as well as multi-mode data, example results of channel semantic characterization are provided and analyzed, which exhibits reasonable and interpretable semantic information.

Towards the Better Ranking Consistency: A Multi-task Learning Framework for Early Stage Ads Ranking

Dividing ads ranking system into retrieval, early, and final stages is a common practice in large scale ads recommendation to balance the efficiency and accuracy. The early stage ranking often uses efficient models to generate candidates out of a set of retrieved ads. The candidates are then fed into a more computationally intensive but accurate final stage ranking system to produce the final ads recommendation. As the early and final stage ranking use different features and model architectures because of system constraints, a serious ranking consistency issue arises where the early stage has a low ads recall, i.e., top ads in the final stage are ranked low in the early stage. In order to pass better ads from the early to the final stage ranking, we propose a multi-task learning framework for early stage ranking to capture multiple final stage ranking components (i.e. ads clicks and ads quality events) and their task relations. With our multi-task learning framework, we can not only achieve serving cost saving from the model consolidation, but also improve the ads recall and ranking consistency. In the online A/B testing, our framework achieves significantly higher click-through rate (CTR), conversion rate (CVR), total value and better ads-quality (e.g. reduced ads cross-out rate) in a large scale industrial ads ranking system.

AdaTT: Adaptive Task-to-Task Fusion Network for Multitask Learning in Recommendations

Multi-task learning (MTL) aims at enhancing the performance and efficiency of machine learning models by training them on multiple tasks simultaneously. However, MTL research faces two challenges: 1) modeling the relationships between tasks to effectively share knowledge between them, and 2) jointly learning task-specific and shared knowledge. In this paper, we present a novel model Adaptive Task-to-Task Fusion Network (AdaTT) to address both challenges. AdaTT is a deep fusion network built with task specific and optional shared fusion units at multiple levels. By leveraging a residual mechanism and gating mechanism for task-to-task fusion, these units adaptively learn shared knowledge and task specific knowledge. To evaluate the performance of AdaTT, we conduct experiments on a public benchmark and an industrial recommendation dataset using various task groups. Results demonstrate AdaTT can significantly outperform existing state-of-the-art baselines.

P$^2$SDF for Neural Indoor Scene Reconstruction

Given only a set of images, neural implicit surface representation has shown its capability in 3D surface reconstruction. However, as the nature of per-scene optimization is based on the volumetric rendering of color, previous neural implicit surface reconstruction methods usually fail in low-textured regions, including the floors, walls, etc., which commonly exist for indoor scenes. Being aware of the fact that these low-textured regions usually correspond to planes, without introducing additional ground-truth supervisory signals or making additional assumptions about the room layout, we propose to leverage a novel Pseudo Plane-regularized Signed Distance Field (P$^2$SDF) for indoor scene reconstruction. Specifically, we consider adjacent pixels with similar colors to be on the same pseudo planes. The plane parameters are then estimated on the fly during training by an efficient and effective two-step scheme. Then the signed distances of the points on the planes are regularized by the estimated plane parameters in the training phase. As the unsupervised plane segments are usually noisy and inaccurate, we propose to assign different weights to the sampled points on the plane in plane estimation as well as the regularization loss. The weights come by fusing the plane segments from different views. As the sampled rays in the planar regions are redundant, leading to inefficient training, we further propose a keypoint-guided rays sampling strategy that attends to the informative textured regions with large color variations, and the implicit network gets a better reconstruction, compared with the original uniform ray sampling strategy. Experiments show that our P$^2$SDF achieves competitive reconstruction performance in Manhattan scenes. Further, as we do not introduce any additional room layout assumption, our P$^2$SDF generalizes well to the reconstruction of non-Manhattan scenes.

A Class-wise Non-salient Region Generalized Framework for Video Semantic Segmentation

Video semantic segmentation (VSS) is beneficial for dealing with dynamic scenes due to the continuous property of the real-world environment. On the one hand, some methods alleviate the predicted inconsistent problem between continuous frames. On the other hand, other methods employ the previous frame as the prior information to assist in segmenting the current frame. Although the previous methods achieve superior performances on the independent and identically distributed (i.i.d) data, they can not generalize well on other unseen domains. Thus, we explore a new task, the video generalizable semantic segmentation (VGSS) task that considers both continuous frames and domain generalization. In this paper, we propose a class-wise non-salient region generalized (CNSG) framework for the VGSS task. Concretely, we first define the class-wise non-salient feature, which describes features of the class-wise non-salient region that carry more generalizable information. Then, we propose a class-wise non-salient feature reasoning strategy to select and enhance the most generalized channels adaptively. Finally, we propose an inter-frame non-salient centroid alignment loss to alleviate the predicted inconsistent problem in the VGSS task. We also extend our video-based framework to the image-based generalizable semantic segmentation (IGSS) task. Experiments demonstrate that our CNSG framework yields significant improvement in the VGSS and IGSS tasks.