11 results on '"Lin, Zinan"'
Search Results
2. Transfer learning for collaborative recommendation with biased and unbiased data
- Author
-
Lin, Zinan, Liu, Dugang, Pan, Weike, Yang, Qiang, and Ming, Zhong
- Published
- 2023
- Full Text
- View/download PDF
3. OMS-DPM: Optimizing the Model Schedule for Diffusion Probabilistic Models
- Author
-
Liu, Enshu, Ning, Xuefei, Lin, Zinan, Yang, Huazhong, and Wang, Yu
- Subjects
FOS: Computer and information sciences ,Computer Science - Machine Learning ,Machine Learning (cs.LG) - Abstract
Diffusion probabilistic models (DPMs) are a new class of generative models that have achieved state-of-the-art generation quality in various domains. Despite the promise, one major drawback of DPMs is the slow generation speed due to the large number of neural network evaluations required in the generation process. In this paper, we reveal an overlooked dimension -- model schedule -- for optimizing the trade-off between generation quality and speed. More specifically, we observe that small models, though having worse generation quality when used alone, could outperform large models in certain generation steps. Therefore, unlike the traditional way of using a single model, using different models in different generation steps in a carefully designed \emph{model schedule} could potentially improve generation quality and speed \emph{simultaneously}. We design OMS-DPM, a predictor-based search algorithm, to optimize the model schedule given an arbitrary generation time budget and a set of pre-trained models. We demonstrate that OMS-DPM can find model schedules that improve generation quality and speed than prior state-of-the-art methods across CIFAR-10, CelebA, ImageNet, and LSUN datasets. When applied to the public checkpoints of the Stable Diffusion model, we are able to accelerate the sampling by 2$\times$ while maintaining the generation quality., Accepted by ICML2023
- Published
- 2023
4. Differentially Private Synthetic Data via Foundation Model APIs 1: Images
- Author
-
Lin, Zinan, Gopi, Sivakanth, Kulkarni, Janardhan, Nori, Harsha, and Yekhanin, Sergey
- Subjects
FOS: Computer and information sciences ,Computer Science - Machine Learning ,Computer Science - Cryptography and Security ,Computer Vision and Pattern Recognition (cs.CV) ,Computer Science - Computer Vision and Pattern Recognition ,Cryptography and Security (cs.CR) ,Machine Learning (cs.LG) - Abstract
Generating differentially private (DP) synthetic data that closely resembles the original private data without leaking sensitive user information is a scalable way to mitigate privacy concerns in the current data-driven world. In contrast to current practices that train customized models for this task, we aim to generate DP Synthetic Data via APIs (DPSDA), where we treat foundation models as blackboxes and only utilize their inference APIs. Such API-based, training-free approaches are easier to deploy as exemplified by the recent surge in the number of API-based apps. These approaches can also leverage the power of large foundation models which are accessible via their inference APIs while the model weights are unreleased. However, this comes with greater challenges due to strictly more restrictive model access and the additional need to protect privacy from the API provider. In this paper, we present a new framework called Private Evolution (PE) to solve this problem and show its initial promise on synthetic images. Surprisingly, PE can match or even outperform state-of-the-art (SOTA) methods without any model training. For example, on CIFAR10 (with ImageNet as the public data), we achieve FID, 38 pages, 33 figures
- Published
- 2023
5. On the Privacy Properties of GAN-generated Samples
- Author
-
Lin, Zinan, Sekar, Vyas, and Fanti, Giulia
- Subjects
FOS: Computer and information sciences ,Condensed Matter::Materials Science ,Computer Science - Machine Learning ,Artificial Intelligence (cs.AI) ,Computer Science - Cryptography and Security ,Computer Science - Artificial Intelligence ,Cryptography and Security (cs.CR) ,Machine Learning (cs.LG) ,Computer Science::Cryptography and Security - Abstract
The privacy implications of generative adversarial networks (GANs) are a topic of great interest, leading to several recent algorithms for training GANs with privacy guarantees. By drawing connections to the generalization properties of GANs, we prove that under some assumptions, GAN-generated samples inherently satisfy some (weak) privacy guarantees. First, we show that if a GAN is trained on m samples and used to generate n samples, the generated samples are (epsilon, delta)-differentially-private for (epsilon, delta) pairs where delta scales as O(n/m). We show that under some special conditions, this upper bound is tight. Next, we study the robustness of GAN-generated samples to membership inference attacks. We model membership inference as a hypothesis test in which the adversary must determine whether a given sample was drawn from the training dataset or from the underlying data distribution. We show that this adversary can achieve an area under the ROC curve that scales no better than O(m^{-1/4})., AISTATS 2021
- Published
- 2022
6. Cooperative regions and partner choice in coded cooperative systems
- Author
-
Lin, Zinan, Erkip, Elza, and Stefanov, Andrej
- Subjects
Telecommunication systems -- Research ,Error-correcting codes -- Analysis - Abstract
User cooperation is an efficient approach to obtain diversity in both centralized and distributed wireless networks. In this paper, we consider a coded cooperative system under quasi-static Rayleigh fading and investigate the partner-choice problem. We find conditions on the interuser and user-to-destination channel qualities for cooperation to be beneficial. Using frame-error rate as a metric, we define the user cooperation gain (G) for evaluating the relative performance improvement of cooperative over direct transmission when a particular channel code is used. We introduce the cooperation decision parameter (CDP), which is a function of user-to-destination average received signal-to-noise ratios (SNRs), and demonstrate that whether cooperation is useful or not (G > 1 or G < 1) depends only on the CDP, not the interuser link quality. We use an analytical formulation of the CDP to investigate user cooperation gain and provide insights on how a user can choose among possible partners to maximize cooperation gain. We first consider the asymptotic performance when one or both partners have high average received SNR at the destination. We then provide conditions on user and destination locations for cooperation to be beneficial for arbitrary SNRs. We illustrate these cooperative regions, and study geometric conditions for the best partner choice. We also define the system cooperation gain and illustrate cooperation benefits for both users. All of our theoretical results are verified through numerical examples. Index Terms--Cooperative diversity, diversity methods, error-correction coding, fading channels, frame-error rate (FER), wireless networks.
- Published
- 2006
7. UCHT-based complex sequences for asynchronous CDMA system
- Author
-
Rahardja, Susanto, Ser, Wee, and Lin, Zinan
- Subjects
Code Division Multiple Access technology - Abstract
The use of orthogonal spreading codes has attracted much attention due to their ability to suppress interference from other users, compared with the nonorthogonal sequences in the synchronous case. In this paper, new sets of orthogonal sequences derived from the unified complex Hadamard transforms (UCHTs) are investigated. Various correlation properties of the sequences are mathematically derived and analyzed. It is shown that some UCHT sequences provide better autocorrelation properties than orthogonal Walsh-Hadamard sequences. Performance comparisons between UCHT sequences, Gold, small set of Kasami, and m-sequences show that some UCHT sequences outperform these well-known spreading sequences under simulation of systems in the presence of multiple access interference and additive white Gaussian noise. Index Terms--Code-division multiple access (CDMA), orthogonal codes, orthogonal transform, spreading sequences, unified complex Hadamard transform (UCHT).
- Published
- 2003
8. Coupling isolated Ni single atoms with sub-10 nm Pd nanocrystals embedded in porous carbon frameworks to boost oxygen electrocatalysis for Zn–air batteries.
- Author
-
Wang, Shangzhi, Lin, Zinan, Li, Mengmeng, Yu, Zehan, Zhang, Minjun, Gong, Mingxing, Tang, Yawen, and Qiu, Xiaoyu
- Abstract
Developing effective bifunctional catalysts for the oxygen reduction and evolution reaction (ORR/OER) is essential for accelerating the cathode efficiency of Zn–air batteries. Herein, atomically dispersed Ni single atoms are supported by sub-10 nm Pd nanocrystals embedded in N-doped carbon frameworks (Ni SAs-Pd@NC), in an effort to achieve superior bifunctional activity in both the ORR and OER. The key synthetic point depends on the protection mechanism of 1-naphthylamine, which could provide a carbon source for Ni SAs and restrict the Pd size under sub-10 nm during 600 °C pyrolysis, simultaneously. The synergistic effect of sub-10 nm Pd with superior ORR activity and Ni–N
4 SAs with favourable OER activity leads to bifunctional catalytic performance, meanwhile the rod-like carbon frameworks with ultrathin, porous and N-doped features contribute to accelerated electron transfer and structural robustness. As a proof-of-concept application, Ni SAs-Pd@NC demonstrates ultrahigh ORR activity with a positive half-wave potential of 0.84 V and a low OER overpotential of 380 mV at 10 mA cm−2 , in an alkaline medium. For a rechargeable Zn–air battery, the Ni SAs-Pd@NC cathode delivers a low charge–discharge voltage gap of 0.87 V, a high energy density of 884.6 W h kgZn −1 , a high power density of 134.2 mW cm−2 and remarkable long-term cyclability for operation over 700 cycles, outperforming commercial Pt/C + RuO2 benchmarks. This work successfully integrates single atom sites with small-sized noble metals to break out their incompatibility in synthesis and to improve their thermostability, which offers a versatile approach to develop single atom-based bifunctional catalysts for energy devices. [ABSTRACT FROM AUTHOR]- Published
- 2022
- Full Text
- View/download PDF
9. Coupling the Atomically Dispersed Fe‐N3 Sites with Sub‐5 nm Pd Nanocrystals Confined in N‐Doped Carbon Nanobelts to Boost the Oxygen Reduction for Microbial Fuel Cells.
- Author
-
Lin, Zinan, Yang, Anzhou, Zhang, Binbin, Liu, Bing, Zhu, Jiawei, Tang, Yawen, and Qiu, Xiaoyu
- Subjects
- *
OXYGEN reduction , *NANOBELTS , *MICROBIAL fuel cells , *NANOCRYSTALS , *FERMI level , *DENSITY functional theory , *MINIMAL surfaces - Abstract
Both the monodispersed Pd/C (2–5 nm) and Fe‐NC single atoms (SAs) are promising non‐Pt catalysts for oxygen reduction reaction (ORR), which belongs to precious metal and nonprecious metal camps, respectively. However, the poles apart of sub‐5 nm Pd/C and Fe‐NC SAs in synthesis and thermostability leave the challenge to integrate them together in one system. Herein, a 1‐naphthylamine protected pyrolysis mechanism is devised to couple the atomically dispersed Fe sites with sub‐5 nm Pd nanocrystals embedded in N‐doped carbon nanobelts (FeN3‐Pd@NC NBs). The FeN3 SAs represent the minimal surface blockage to tune the electronic structure of Pd, while the carbon frameworks are born with ultrathin, porous, and N‐doped feature's. As inspired, the FeN3‐Pd@NC NBs exhibit outstanding activity (E1/2 = 0.926 V) and durability (2 mV decay in E1/2 after 2000 cycles) for ORR, as well as achieving a maximum power density of 831.2 mW cm−2 in a microbial fuel cell operated for over 100 d. Density functional theory calculation reveals that the FeN3 SAs can shift the density of states of Pd toward the Fermi level, and their coupling can decrease the limiting reaction barrier with a value of −0.62 eV, thus greatly accelerating the ORR kinetics. [ABSTRACT FROM AUTHOR]
- Published
- 2022
- Full Text
- View/download PDF
10. InfoGAN-CR and ModelCentrality: Self-supervised Model Training and Selection for Disentangling GANs
- Author
-
Lin, Zinan, Thekumparampil, Kiran Koshy, Fanti, Giulia, and Oh, Sewoong
- Subjects
FOS: Computer and information sciences ,Computer Science - Machine Learning ,ComputingMethodologies_PATTERNRECOGNITION ,Statistics - Machine Learning ,Machine Learning (stat.ML) ,Machine Learning (cs.LG) - Abstract
Disentangled generative models map a latent code vector to a target space, while enforcing that a subset of the learned latent codes are interpretable and associated with distinct properties of the target distribution. Recent advances have been dominated by Variational AutoEncoder (VAE)-based methods, while training disentangled generative adversarial networks (GANs) remains challenging. In this work, we show that the dominant challenges facing disentangled GANs can be mitigated through the use of self-supervision. We make two main contributions: first, we design a novel approach for training disentangled GANs with self-supervision. We propose contrastive regularizer, which is inspired by a natural notion of disentanglement: latent traversal. This achieves higher disentanglement scores than state-of-the-art VAE- and GAN-based approaches. Second, we propose an unsupervised model selection scheme called ModelCentrality, which uses generated synthetic samples to compute the medoid (multi-dimensional generalization of median) of a collection of models. The current common practice of hyper-parameter tuning requires using ground-truths samples, each labelled with known perfect disentangled latent codes. As real datasets are not equipped with such labels, we propose an unsupervised model selection scheme and show that it finds a model close to the best one, for both VAEs and GANs. Combining contrastive regularization with ModelCentrality, we improve upon the state-of-the-art disentanglement scores significantly, without accessing the supervised data., Published in ICML 2020. 45 pages, 52 figures, a new unsupervised model selection scheme (ModelCentrality) is introduced in this version
- Published
- 2019
11. A comparison of MAC aggregation Vs. PHY bonding for WLANs in TV White Spaces.
- Author
-
Lin, Zinan, Ghosh, Monisha, and Demir, Alpaslan
- Published
- 2013
- Full Text
- View/download PDF
Catalog
Discovery Service for Jio Institute Digital Library
For full access to our library's resources, please sign in.