Search

There have been a plethora of research on multi-level memory devices, where the resistive random-access memory (RRAM) is a prominent example. Although it is easy to write an RRAM device into multiple (even quasi-continuous) states, it suffers from the inherent variations that should limit the storage capacity, especially in the open-loop writing scenario. There have been many experimental results in this regard, however, it lacks a comprehensive analysis of the valid multi-bit storage capability, especially in theoretical terms. The absence of such an insight usually results in misleading conclusions that either exaggerate or underestimate the storage capacity of RRAM devices. Here, by the concept of information theory, we present a model for evaluating the storage capacity of open-loop written RRAM. Based on the experimental results in the literature and the test results of our own devices, we have carefully examined the effects of number of pre-defined levels, conductance variation, and conductance range, on the storage capacity. The analysis leads to a conclusion that the maximum capacity of RRAM devices is around 4 bits., Comment: This work has been presented at the conference ICSICT 2024

The locate-then-edit paradigm has shown significant promise for knowledge editing (KE) in Large Language Models (LLMs). While previous methods perform well on single-hop fact recall tasks, they consistently struggle with multi-hop factual recall tasks involving newly edited knowledge. In this paper, leveraging tools in mechanistic interpretability, we first identify that in multi-hop tasks, LLMs tend to retrieve implicit subject knowledge from deeper MLP layers, unlike single-hop tasks, which rely on earlier layers. This distinction explains the poor performance of current methods in multi-hop queries, as they primarily focus on editing shallow layers, leaving deeper layers unchanged. To address this, we propose IFMET, a novel locate-then-edit KE approach designed to edit both shallow and deep MLP layers. IFMET employs multi-hop editing prompts and supplementary sets to locate and modify knowledge across different reasoning stages. Experimental results demonstrate that IFMET significantly improves performance on multi-hop factual recall tasks, effectively overcoming the limitations of previous locate-then-edit methods., Comment: 21 pages

Video Moment Retrieval, which aims to locate in-context video moments according to a natural language query, is an essential task for cross-modal grounding. Existing methods focus on enhancing the cross-modal interactions between all moments and the textual description for video understanding. However, constantly interacting with all locations is unreasonable because of uneven semantic distribution across the timeline and noisy visual backgrounds. This paper proposes a cross-modal Context Denoising Network (CDNet) for accurate moment retrieval by disentangling complex correlations and denoising irrelevant dynamics.Specifically, we propose a query-guided semantic disentanglement (QSD) to decouple video moments by estimating alignment levels according to the global and fine-grained correlation. A Context-aware Dynamic Denoisement (CDD) is proposed to enhance understanding of aligned spatial-temporal details by learning a group of query-relevant offsets. Extensive experiments on public benchmarks demonstrate that the proposed CDNet achieves state-of-the-art performances.

Large Language Models (LLMs) represent a significant stride toward Artificial General Intelligence. As scaling laws underscore the potential of increasing model sizes, the academic community has intensified its investigations into LLMs with capacities exceeding 50 billion parameters. This technical report builds on our prior work with Tele-FLM (also known as FLM-2), a publicly available 52-billion-parameter model. We delve into two primary areas: we first discuss our observation of Supervised Fine-tuning (SFT) on Tele-FLM-52B, which supports the "less is more" approach for SFT data construction; second, we demonstrate our experiments and analyses on the best practices for progressively growing a model from 52 billion to 102 billion, and subsequently to 1 trillion parameters. We will open-source a 1T model checkpoint, namely Tele-FLM-1T, to advance further training and research., Comment: For the Tele-FLM-52B tech report, see also 2404.16645

Large language models (LLMs) have showcased profound capabilities in language understanding and generation, facilitating a wide array of applications. However, there is a notable paucity of detailed, open-sourced methodologies on efficiently scaling LLMs beyond 50 billion parameters with minimum trial-and-error cost and computational resources. In this report, we introduce Tele-FLM (aka FLM-2), a 52B open-sourced multilingual large language model that features a stable, efficient pre-training paradigm and enhanced factual judgment capabilities. Tele-FLM demonstrates superior multilingual language modeling abilities, measured by BPB on textual corpus. Besides, in both English and Chinese foundation model evaluation, it is comparable to strong open-sourced models that involve larger pre-training FLOPs, such as Llama2-70B and DeepSeek-67B. In addition to the model weights, we share the core designs, engineering practices, and training details, which we expect to benefit both the academic and industrial communities.

Text-video retrieval aims to find the most relevant cross-modal samples for a given query. Recent methods focus on modeling the whole spatial-temporal relations. However, since video clips contain more diverse content than captions, the model aligning these asymmetric video-text pairs has a high risk of retrieving many false positive results. In this paper, we propose Probabilistic Token Aggregation (ProTA) to handle cross-modal interaction with content asymmetry. Specifically, we propose dual partial-related aggregation to disentangle and re-aggregate token representations in both low-dimension and high-dimension spaces. We propose token-based probabilistic alignment to generate token-level probabilistic representation and maintain the feature representation diversity. In addition, an adaptive contrastive loss is proposed to learn compact cross-modal distribution space. Based on extensive experiments, ProTA achieves significant improvements on MSR-VTT (50.9%), LSMDC (25.8%), and DiDeMo (47.2%).

In our dynamic world where data arrives in a continuous stream, continual learning enables us to incrementally add new tasks/domains without the need to retrain from scratch. A major challenge in continual learning of language model is catastrophic forgetting, the tendency of models to forget knowledge from previously trained tasks/domains when training on new ones. This paper studies dialog generation under the continual learning setting. We propose a novel method that 1) uses \textit{Text-Mixup} as data augmentation to avoid model overfitting on replay memory and 2) leverages Batch-Nuclear Norm Maximization (BNNM) to alleviate the problem of mode collapse. Experiments on a $37$-domain task-oriented dialog dataset and DailyDialog (a $10$-domain chitchat dataset) demonstrate that our proposed approach outperforms the state-of-the-art in continual learning., Comment: 12 pages, 2 figures

In this technical report, we present TeleChat, a collection of large language models (LLMs) with parameters of 3 billion, 7 billion and 12 billion. It includes pretrained language models as well as fine-tuned chat models that is aligned with human preferences. TeleChat is initially pretrained on an extensive corpus containing a diverse collection of texts from both English and Chinese languages, including trillions of tokens. Subsequently, the model undergoes fine-tuning to align with human preferences, following a detailed methodology that we describe. We evaluate the performance of TeleChat on various tasks, including language understanding, mathematics, reasoning, code generation, and knowledge-based question answering. Our findings indicate that TeleChat achieves comparable performance to other open-source models of similar size across a wide range of public benchmarks. To support future research and applications utilizing LLMs, we release the fine-tuned model checkpoints of TeleChat's 7B and 12B variant, along with code and a portion of our pretraining data, to the public community., Comment: 28 pages, 2 figures

Defects in atomically thin materials can drive new functionalities and expand applications to multifunctional systems that are monolithically integrated. An ability to control formation of defects during the synthesis process is an important capability to create practical deployment opportunities. Molybdenum disulfide (MoS$_2$), a two-dimensional (2D) semiconducting material harbors intrinsic defects that can be harnessed to achieve tuneable electronic, optoelectronic, and electrochemical devices. However, achieving precise control over defect formation within monolayer MoS$_2$, while maintaining the structural integrity of the crystals remains a notable challenge. Here, we present a one-step, in-situ defect engineering approach for monolayer MoS$_2$ using a pressure dependent chemical vapour deposition (CVD) process. Monolayer MoS$_2$ grown in low-pressure CVD conditions (LP-MoS$_2$) produces sulfur vacancy (Vs) induced defect rich crystals primarily attributed to the kinetics of the growth conditions. Conversely, atmospheric pressure CVD grown MoS$_2$ (AP-MoS$_2$) passivates these Vs defects with oxygen. This disparity in defect profiles profoundly impacts crucial functional properties and device performance. AP-MoS$_2$ shows a drastically enhanced photoluminescence, which is significantly quenched in LP-MoS$_2$ attributed to in-gap electron donor states induced by the Vs defects. However, the n-doping induced by the Vs defects in LP-MoS$_2$ generates enhanced photoresponsivity and detectivity in our fabricated photodetectors compared to the AP-MoS$_2$ based devices. Defect-rich LP-MoS$_2$ outperforms AP-MoS$_2$ as channel layers of field-effect transistors (FETs), as well as electrocatalytic material for hydrogen evolution reaction (HER). This work presents a single-step CVD approach for in-situ defect engineering in monolayer MoS$_2$ and presents a pathway to control defects in other monolayer material systems., Comment: 29 pages, 5 figures

Dictionary learning is an effective tool for pattern recognition and classification of time series data. Among various dictionary learning techniques, the dynamic time warping (DTW) is commonly used for dealing with temporal delays, scaling, transformation, and many other kinds of temporal misalignments issues. However, the DTW suffers overfitting or information loss due to its discrete nature in aligning time series data. To address this issue, we propose a generalized time warping invariant dictionary learning algorithm in this paper. Our approach features a generalized time warping operator, which consists of linear combinations of continuous basis functions for facilitating continuous temporal warping. The integration of the proposed operator and the dictionary learning is formulated as an optimization problem, where the block coordinate descent method is employed to jointly optimize warping paths, dictionaries, and sparseness coefficients. The optimized results are then used as hyperspace distance measures to feed classification and clustering algorithms. The superiority of the proposed method in terms of dictionary learning, classification, and clustering is validated through ten sets of public datasets in comparing with various benchmark methods.

The periodic Gaussian process (PGP) has been increasingly used to model periodic data due to its high accuracy. Yet, computing the likelihood of PGP has a high computational complexity of $\mathcal{O}\left(n^{3}\right)$ ($n$ is the data size), which hinders its wide application. To address this issue, we propose a novel circulant PGP (CPGP) model for large-scale periodic data collected at grids that are commonly seen in signal processing applications. The proposed CPGP decomposes the log-likelihood of PGP into the sum of two computationally scalable composite log-likelihoods, which do not involve any approximations. Computing the likelihood of CPGP requires only $\mathcal{O}\left(p^{2}\right)$ (or $\mathcal{O}\left(p\log p\right)$ in some special cases) time for grid observations, where the segment length $p$ is independent of and much smaller than $n$. Simulations and real case studies are presented to show the superiority of CPGP over some state-of-the-art methods, especially for applications requiring periodicity estimation. This new modeling technique can greatly advance the applicability of PGP in many areas and allow the modeling of many previously intractable problems., Comment: Accepted for publication in Technometrics