Da-Cheng Juan

Luyu Gao, Zhuyun Dai, Panupong Pasupat et al. · cmu

Language models (LMs) now excel at many tasks such as few-shot learning, question answering, reasoning, and dialog. However, they sometimes generate unsupported or misleading content. A user cannot easily determine whether their outputs are trustworthy or not, because most LMs do not have any built-in mechanism for attribution to external evidence. To enable attribution while still preserving all the powerful advantages of recent generation models, we propose RARR (Retrofit Attribution using Research and Revision), a system that 1) automatically finds attribution for the output of any text generation model and 2) post-edits the output to fix unsupported content while preserving the original output as much as possible. When applied to the output of several state-of-the-art LMs on a diverse set of generation tasks, we find that RARR significantly improves attribution while otherwise preserving the original input to a much greater degree than previously explored edit models. Furthermore, the implementation of RARR requires only a handful of training examples, a large language model, and standard web search.

Yu-Jen Chen, Wei-Hsiang Shen, Hao-Wei Chung et al. · pku

Medical report generation is a challenging task since it is time-consuming and requires expertise from experienced radiologists. The goal of medical report generation is to accurately capture and describe the image findings. Previous works pretrain their visual encoding neural networks with large datasets in different domains, which cannot learn general visual representation in the specific medical domain. In this work, we propose a medical report generation framework that uses a contrastive learning approach to pretrain the visual encoder and requires no additional meta information. In addition, we adopt lung segmentation as an augmentation method in the contrastive learning framework. This segmentation guides the network to focus on encoding the visual feature within the lung region. Experimental results show that the proposed framework improves the performance and the quality of the generated medical reports both quantitatively and qualitatively.

Zhenting Qi, Susanna Maria Baby, Stefanie Anna Baby et al.

Recent work suggests that large language models (LLMs) can improve through self-evolution (SE), using supervision signals generated by the model itself. In this work, we ask: under a strict closed-loop setup, where the self-evolution algorithm has access only to an unlabeled prompt set and a base model, how close can internally generated supervision come to oracle-supervised training? We analyze four representative strategies in a unified offline self-evolution framework: single-round verification, multi-turn revision with feedback, iterative training, and curriculum learning. Our primary experiments use Knights and Knaves (KK) logical reasoning tasks, which provide deterministic solutions, controlled difficulty levels, and a clean testbed for easy-to-hard generalization. We first show that self-evolution consistently improves over the base model, but plateaus after excessive training compute is invested, and eventually still leaves a non-trivial gap to oracle supervision. We find that multi-turn critic-revision with large models can reach strong self-evolution performance, with Gemma 12B nearly matching oracle-supervised training. Beyond Knights and Knaves, we also evaluate self-evolution on real-world reasoning benchmarks, where gains are also modest. Overall, our results characterize when closed-loop self-evolution can help and show how internally generated supervision remains insufficient under this minimal formulation.

Cyrus Rashtchian, Charles Herrmann, Chun-Sung Ferng et al.

Probes are small networks that predict properties of underlying data from embeddings, and they provide a targeted, effective way to illuminate the information contained in embeddings. While analysis through the use of probes has become standard in NLP, there has been much less exploration in vision. Image foundation models have primarily been evaluated for semantic content. Better understanding the non-semantic information in popular embeddings (e.g., MAE, SimCLR, or CLIP) will shed new light both on the training algorithms and on the uses for these foundation models. We design a systematic transformation prediction task and measure the visual content of embeddings along many axes, including image style, quality, and a range of natural and artificial transformations. Surprisingly, six embeddings (including SimCLR) encode enough non-semantic information to identify dozens of transformations. We also consider a generalization task, where we group similar transformations and hold out several for testing. We find that image-text models (CLIP and ALIGN) are better at recognizing new examples of style transfer than masking-based models (CAN and MAE). Overall, our results suggest that the choice of pre-training algorithm impacts the types of information in the embedding, and certain models are better than others for non-semantic downstream tasks.

Jiao Sun, Deqing Fu, Yushi Hu et al.

Despite their wide-spread success, Text-to-Image models (T2I) still struggle to produce images that are both aesthetically pleasing and faithful to the user's input text. We introduce DreamSync, a model-agnostic training algorithm by design that improves T2I models to be faithful to the text input. DreamSync builds off a recent insight from TIFA's evaluation framework -- that large vision-language models (VLMs) can effectively identify the fine-grained discrepancies between generated images and the text inputs. DreamSync uses this insight to train T2I models without any labeled data; it improves T2I models using its own generations. First, it prompts the model to generate several candidate images for a given input text. Then, it uses two VLMs to select the best generation: a Visual Question Answering model that measures the alignment of generated images to the text, and another that measures the generation's aesthetic quality. After selection, we use LoRA to iteratively finetune the T2I model to guide its generation towards the selected best generations. DreamSync does not need any additional human annotation. model architecture changes, or reinforcement learning. Despite its simplicity, DreamSync improves both the semantic alignment and aesthetic appeal of two diffusion-based T2I models, evidenced by multiple benchmarks (+1.7% on TIFA, +2.9% on DSG1K, +3.4% on VILA aesthetic) and human evaluation.

Yicheng Fan, Dana Alon, Jingyue Shen et al.

Neural Architecture Search (NAS) has become a popular method for discovering effective model architectures, especially for target hardware. As such, NAS methods that find optimal architectures under constraints are essential. In our paper, we propose LayerNAS to address the challenge of multi-objective NAS by transforming it into a combinatorial optimization problem, which effectively constrains the search complexity to be polynomial. For a model architecture with $L$ layers, we perform layerwise-search for each layer, selecting from a set of search options $\mathbb{S}$. LayerNAS groups model candidates based on one objective, such as model size or latency, and searches for the optimal model based on another objective, thereby splitting the cost and reward elements of the search. This approach limits the search complexity to $ O(H \cdot |\mathbb{S}| \cdot L) $, where $H$ is a constant set in LayerNAS. Our experiments show that LayerNAS is able to consistently discover superior models across a variety of search spaces in comparison to strong baselines, including search spaces derived from NATS-Bench, MobileNetV2 and MobileNetV3.

Gheorghe Comanici, Eric Bieber, Mike Schaekermann et al. · amazon-science, baidu

In this report, we introduce the Gemini 2.X model family: Gemini 2.5 Pro and Gemini 2.5 Flash, as well as our earlier Gemini 2.0 Flash and Flash-Lite models. Gemini 2.5 Pro is our most capable model yet, achieving SoTA performance on frontier coding and reasoning benchmarks. In addition to its incredible coding and reasoning skills, Gemini 2.5 Pro is a thinking model that excels at multimodal understanding and it is now able to process up to 3 hours of video content. Its unique combination of long context, multimodal and reasoning capabilities can be combined to unlock new agentic workflows. Gemini 2.5 Flash provides excellent reasoning abilities at a fraction of the compute and latency requirements and Gemini 2.0 Flash and Flash-Lite provide high performance at low latency and cost. Taken together, the Gemini 2.X model generation spans the full Pareto frontier of model capability vs cost, allowing users to explore the boundaries of what is possible with complex agentic problem solving.

Chun-Ta Lu, Yun Zeng, Da-Cheng Juan et al.

In this work, we propose CARLS, a novel framework for augmenting the capacity of existing deep learning frameworks by enabling multiple components -- model trainers, knowledge makers and knowledge banks -- to concertedly work together in an asynchronous fashion across hardware platforms. The proposed CARLS is particularly suitable for learning paradigms where model training benefits from additional knowledge inferred or discovered during training, such as node embeddings for graph neural networks or reliable pseudo labels from model predictions. We also describe three learning paradigms -- semi-supervised learning, curriculum learning and multimodal learning -- as examples that can be scaled up efficiently by CARLS. One version of CARLS has been open-sourced and available for download at: https://github.com/tensorflow/neural-structured-learning/tree/master/research/carls

Hailey Joren, Jianyi Zhang, Chun-Sung Ferng et al.

Augmenting LLMs with context leads to improved performance across many applications. Despite much research on Retrieval Augmented Generation (RAG) systems, an open question is whether errors arise because LLMs fail to utilize the context from retrieval or the context itself is insufficient to answer the query. To shed light on this, we develop a new notion of sufficient context, along with a method to classify instances that have enough information to answer the query. We then use sufficient context to analyze several models and datasets. By stratifying errors based on context sufficiency, we find that larger models with higher baseline performance (Gemini 1.5 Pro, GPT 4o, Claude 3.5) excel at answering queries when the context is sufficient, but often output incorrect answers instead of abstaining when the context is not. On the other hand, smaller models with lower baseline performance (Mistral 3, Gemma 2) hallucinate or abstain often, even with sufficient context. We further categorize cases when the context is useful, and improves accuracy, even though it does not fully answer the query and the model errs without the context. Building on our findings, we explore ways to reduce hallucinations in RAG systems, including a new selective generation method that leverages sufficient context information for guided abstention. Our method improves the fraction of correct answers among times where the model responds by 2--10\% for Gemini, GPT, and Gemma. Key findings and the prompts used in our autorater analysis are available on our github.

Jianyi Zhang, Da-Cheng Juan, Cyrus Rashtchian et al.

Large language models (LLMs) have demonstrated remarkable capabilities, but their outputs can sometimes be unreliable or factually incorrect. To address this, we introduce Self Logits Evolution Decoding (SLED), a novel decoding framework that enhances the truthfulness of LLMs without relying on external knowledge bases or requiring further fine-tuning. From an optimization perspective, our SLED framework leverages the latent knowledge embedded within the LLM by contrasting the output logits from the final layer with those from early layers. It then utilizes an approximate gradient approach to enable latent knowledge to guide the self-refinement of outputs, thereby effectively improving factual accuracy. Extensive experiments have been conducted on established benchmarks across a diverse range of model families (Gemma, Qwen, Mixtral, gpt-oss) and scales (from 1B to 45B), including more advanced architectural configurations such as the mixture of experts (MoE). Our evaluation spans a wide variety of tasks and the results demonstrate that SLED consistently improves factual accuracy compared to existing decoding methods while maintaining natural language fluency and negligible latency overhead. Furthermore, it can be flexibly combined with other decoding methods to further enhance their performance.

Ko-Wei Huang, Yi-Fu Fu, Ching-Yu Tsai et al.

We investigate how Large Language Models (LLMs) distinguish between memorization and generalization at the neuron level. Through carefully designed tasks, we identify distinct neuron subsets responsible for each behavior. Experiments on both a GPT-2 model trained from scratch and a pretrained LLaMA-3.2 model fine-tuned with LoRA show consistent neuron-level specialization. We further demonstrate that inference-time interventions on these neurons can steer the model's behavior toward memorization or generalization. To assess robustness, we evaluate intra-task and inter-task consistency, confirming that these neuron-behavior associations reflect generalizable patterns rather than dataset-specific artifacts. Our findings reveal modular structure in LLMs and enable controlling memorization and generalization behaviors at inference time.

Gemini Team, Rohan Anil, Sebastian Borgeaud et al.

This report introduces a new family of multimodal models, Gemini, that exhibit remarkable capabilities across image, audio, video, and text understanding. The Gemini family consists of Ultra, Pro, and Nano sizes, suitable for applications ranging from complex reasoning tasks to on-device memory-constrained use-cases. Evaluation on a broad range of benchmarks shows that our most-capable Gemini Ultra model advances the state of the art in 30 of 32 of these benchmarks - notably being the first model to achieve human-expert performance on the well-studied exam benchmark MMLU, and improving the state of the art in every one of the 20 multimodal benchmarks we examined. We believe that the new capabilities of the Gemini family in cross-modal reasoning and language understanding will enable a wide variety of use cases. We discuss our approach toward post-training and deploying Gemini models responsibly to users through services including Gemini, Gemini Advanced, Google AI Studio, and Cloud Vertex AI.

Wei-Cheng Tseng, Wei Wei, Da-Cheng Juan et al.

Designing an effective communication mechanism among agents in reinforcement learning has been a challenging task, especially for real-world applications. The number of agents can grow or an environment sometimes needs to interact with a changing number of agents in real-world scenarios. To this end, a multi-agent framework needs to handle various scenarios of agents, in terms of both scales and dynamics, for being practical to real-world applications. We formulate the multi-agent environment with a different number of agents as a multi-tasking problem and propose a meta reinforcement learning (meta-RL) framework to tackle this problem. The proposed framework employs a meta-learned Communication Pattern Recognition (CPR) module to identify communication behavior and extract information that facilitates the training process. Experimental results are poised to demonstrate that the proposed framework (a) generalizes to an unseen larger number of agents and (b) allows the number of agents to change between episodes. The ablation study is also provided to reason the proposed CPR design and show such design is effective.

Zheyu Yan, Da-Cheng Juan, Xiaobo Sharon Hu et al.

Emerging device-based Computing-in-memory (CiM) has been proved to be a promising candidate for high-energy efficiency deep neural network (DNN) computations. However, most emerging devices suffer uncertainty issues, resulting in a difference between actual data stored and the weight value it is designed to be. This leads to an accuracy drop from trained models to actually deployed platforms. In this work, we offer a thorough analysis of the effect of such uncertainties-induced changes in DNN models. To reduce the impact of device uncertainties, we propose UAE, an uncertainty-aware Neural Architecture Search scheme to identify a DNN model that is both accurate and robust against device uncertainties.

Yi Tay, Mostafa Dehghani, Vamsi Aribandi et al.

This paper proposes Omnidirectional Representations from Transformers (OmniNet). In OmniNet, instead of maintaining a strictly horizontal receptive field, each token is allowed to attend to all tokens in the entire network. This process can also be interpreted as a form of extreme or intensive attention mechanism that has the receptive field of the entire width and depth of the network. To this end, the omnidirectional attention is learned via a meta-learner, which is essentially another self-attention based model. In order to mitigate the computationally expensive costs of full receptive field attention, we leverage efficient self-attention models such as kernel-based (Choromanski et al.), low-rank attention (Wang et al.) and/or Big Bird (Zaheer et al.) as the meta-learner. Extensive experiments are conducted on autoregressive language modeling (LM1B, C4), Machine Translation, Long Range Arena (LRA), and Image Recognition. The experiments show that OmniNet achieves considerable improvements across these tasks, including achieving state-of-the-art performance on LM1B, WMT'14 En-De/En-Fr, and Long Range Arena. Moreover, using omnidirectional representation in Vision Transformers leads to significant improvements on image recognition tasks on both few-shot learning and fine-tuning setups.

Shuai Zhang, Yi Tay, Wenqi Jiang et al.

Learning embedding spaces of suitable geometry is critical for representation learning. In order for learned representations to be effective and efficient, it is ideal that the geometric inductive bias aligns well with the underlying structure of the data. In this paper, we propose Switch Spaces, a data-driven approach for learning representations in product space. Specifically, product spaces (or manifolds) are spaces of mixed curvature, i.e., a combination of multiple euclidean and non-euclidean (hyperbolic, spherical) manifolds. To this end, we introduce sparse gating mechanisms that learn to choose, combine and switch spaces, allowing them to be switchable depending on the input data with specialization. Additionally, the proposed method is also efficient and has a constant computational complexity regardless of the model size. Experiments on knowledge graph completion and item recommendations show that the proposed switch space achieves new state-of-the-art performances, outperforming pure product spaces and recently proposed task-specific models.

Jia Li, Tomas Yu, Da-Cheng Juan et al.

Recent studies have indicated that Graph Convolutional Networks (GCNs) act as a \emph{low pass} filter in spectral domain and encode smoothed node representations. In this paper, we consider their opposite, namely Graph Deconvolutional Networks (GDNs) that reconstruct graph signals from smoothed node representations. We motivate the design of Graph Deconvolutional Networks via a combination of inverse filters in spectral domain and de-noising layers in wavelet domain, as the inverse operation results in a \emph{high pass} filter and may amplify the noise. Based on the proposed GDN, we further propose a graph autoencoder framework that first encodes smoothed graph representations with GCN and then decodes accurate graph signals with GDN. We demonstrate the effectiveness of the proposed method on several tasks including unsupervised graph-level representation , social recommendation and graph generation

Pranjal Awasthi, George Yu, Chun-Sung Ferng et al.

Adversarial robustness corresponds to the susceptibility of deep neural networks to imperceptible perturbations made at test time. In the context of image tasks, many algorithms have been proposed to make neural networks robust to adversarial perturbations made to the input pixels. These perturbations are typically measured in an $\ell_p$ norm. However, robustness often holds only for the specific attack used for training. In this work we extend the above setting to consider the problem of training of deep neural networks that can be made simultaneously robust to perturbations applied in multiple natural representation spaces. For the case of image data, examples include the standard pixel representation as well as the representation in the discrete cosine transform~(DCT) basis. We design a theoretically sound algorithm with formal guarantees for the above problem. Furthermore, our guarantees also hold when the goal is to require robustness with respect to multiple $\ell_p$ norm based attacks. We then derive an efficient practical implementation and demonstrate the effectiveness of our approach on standard datasets for image classification.

Yi Tay, Zhe Zhao, Dara Bahri et al.

Achieving state-of-the-art performance on natural language understanding tasks typically relies on fine-tuning a fresh model for every task. Consequently, this approach leads to a higher overall parameter cost, along with higher technical maintenance for serving multiple models. Learning a single multi-task model that is able to do well for all the tasks has been a challenging and yet attractive proposition. In this paper, we propose \textsc{HyperGrid}, a new approach for highly effective multi-task learning. The proposed approach is based on a decomposable hypernetwork that learns grid-wise projections that help to specialize regions in weight matrices for different tasks. In order to construct the proposed hypernetwork, our method learns the interactions and composition between a global (task-agnostic) state and a local task-specific state. We apply our proposed \textsc{HyperGrid} on the current state-of-the-art T5 model, demonstrating strong performance across the GLUE and SuperGLUE benchmarks when using only a single multi-task model. Our method helps bridge the gap between fine-tuning and multi-task learning approaches.

Hsin-Ping Chou, Shih-Chieh Chang, Jia-Yu Pan et al.

Deep image classifiers often perform poorly when training data are heavily class-imbalanced. In this work, we propose a new regularization technique, Remix, that relaxes Mixup's formulation and enables the mixing factors of features and labels to be disentangled. Specifically, when mixing two samples, while features are mixed in the same fashion as Mixup, Remix assigns the label in favor of the minority class by providing a disproportionately higher weight to the minority class. By doing so, the classifier learns to push the decision boundaries towards the majority classes and balance the generalization error between majority and minority classes. We have studied the state-of-the art regularization techniques such as Mixup, Manifold Mixup and CutMix under class-imbalanced regime, and shown that the proposed Remix significantly outperforms these state-of-the-arts and several re-weighting and re-sampling techniques, on the imbalanced datasets constructed by CIFAR-10, CIFAR-100, and CINIC-10. We have also evaluated Remix on a real-world large-scale imbalanced dataset, iNaturalist 2018. The experimental results confirmed that Remix provides consistent and significant improvements over the previous methods.

Li-Huang Tsai, Shih-Chieh Chang, Yu-Ting Chen et al.

Analog computing hardwares, such as Processing-in-memory (PIM) accelerators, have gradually received more attention for accelerating the neural network computations. However, PIM accelerators often suffer from intrinsic noise in the physical components, making it challenging for neural network models to achieve the same performance as on the digital hardware. Previous works in mitigating intrinsic noise assumed the knowledge of the noise model, and retraining the neural networks accordingly was required. In this paper, we propose a noise-agnostic method to achieve robust neural network performance against any noise setting. Our key observation is that the degradation of performance is due to the distribution shifts in network activations, which are caused by the noise. To properly track the shifts and calibrate the biased distributions, we propose a "noise-aware" batch normalization layer, which is able to align the distributions of the activations under variational noise inherent in the analog environments. Our method is simple, easy to implement, general to various noise settings, and does not need to retrain the models. We conduct experiments on several tasks in computer vision, including classification, object detection and semantic segmentation. The results demonstrate the effectiveness of our method, achieving robust performance under a wide range of noise settings, more reliable than existing methods. We believe that our simple yet general method can facilitate the adoption of analog computing devices for neural networks.

Yi Tay, Dara Bahri, Donald Metzler et al.

The dot product self-attention is known to be central and indispensable to state-of-the-art Transformer models. But is it really required? This paper investigates the true importance and contribution of the dot product-based self-attention mechanism on the performance of Transformer models. Via extensive experiments, we find that (1) random alignment matrices surprisingly perform quite competitively and (2) learning attention weights from token-token (query-key) interactions is useful but not that important after all. To this end, we propose \textsc{Synthesizer}, a model that learns synthetic attention weights without token-token interactions. In our experiments, we first show that simple Synthesizers achieve highly competitive performance when compared against vanilla Transformer models across a range of tasks, including machine translation, language modeling, text generation and GLUE/SuperGLUE benchmarks. When composed with dot product attention, we find that Synthesizers consistently outperform Transformers. Moreover, we conduct additional comparisons of Synthesizers against Dynamic Convolutions, showing that simple Random Synthesizer is not only $60\%$ faster but also improves perplexity by a relative $3.5\%$. Finally, we show that simple factorized Synthesizers can outperform Linformers on encoding only tasks.

Ines Chami, Adva Wolf, Da-Cheng Juan et al.

Knowledge graph (KG) embeddings learn low-dimensional representations of entities and relations to predict missing facts. KGs often exhibit hierarchical and logical patterns which must be preserved in the embedding space. For hierarchical data, hyperbolic embedding methods have shown promise for high-fidelity and parsimonious representations. However, existing hyperbolic embedding methods do not account for the rich logical patterns in KGs. In this work, we introduce a class of hyperbolic KG embedding models that simultaneously capture hierarchical and logical patterns. Our approach combines hyperbolic reflections and rotations with attention to model complex relational patterns. Experimental results on standard KG benchmarks show that our method improves over previous Euclidean- and hyperbolic-based efforts by up to 6.1% in mean reciprocal rank (MRR) in low dimensions. Furthermore, we observe that different geometric transformations capture different types of relations while attention-based transformations generalize to multiple relations. In high dimensions, our approach yields new state-of-the-art MRRs of 49.6% on WN18RR and 57.7% on YAGO3-10.

Yi Tay, Dara Bahri, Liu Yang et al.

We propose Sparse Sinkhorn Attention, a new efficient and sparse method for learning to attend. Our method is based on differentiable sorting of internal representations. Concretely, we introduce a meta sorting network that learns to generate latent permutations over sequences. Given sorted sequences, we are then able to compute quasi-global attention with only local windows, improving the memory efficiency of the attention module. To this end, we propose new algorithmic innovations such as Causal Sinkhorn Balancing and SortCut, a dynamic sequence truncation method for tailoring Sinkhorn Attention for encoding and/or decoding purposes. Via extensive experiments on algorithmic seq2seq sorting, language modeling, pixel-wise image generation, document classification and natural language inference, we demonstrate that our memory efficient Sinkhorn Attention method is competitive with vanilla attention and consistently outperforms recently proposed efficient Transformer models such as Sparse Transformers.

Hao-Yun Chen, Li-Huang Tsai, Shih-Chieh Chang et al.

Label hierarchies widely exist in many vision-related problems, ranging from explicit label hierarchies existed in image classification to latent label hierarchies existed in semantic segmentation. Nevertheless, state-of-the-art methods often deploy cross-entropy loss that implicitly assumes class labels to be exclusive and thus independence from each other. Motivated by the fact that classes from the same parental category usually share certain similarity, we design a new training diagram called Hierarchical Complement Objective Training (HCOT) that leverages the information from label hierarchy. HCOT maximizes the probability of the ground truth class, and at the same time, neutralizes the probabilities of rest of the classes in a hierarchical fashion, making the model take advantage of the label hierarchy explicitly. The proposed HCOT is evaluated on both image classification and semantic segmentation tasks. Experimental results confirm that HCOT outperforms state-of-the-art models in CIFAR-100, ImageNet-2012, and PASCAL-Context. The study further demonstrates that HCOT can be applied on tasks with latent label hierarchies, which is a common characteristic in many machine learning tasks.

Yu-Lun Hsieh, Minhao Cheng, Da-Cheng Juan et al.

This work proposes a novel algorithm to generate natural language adversarial input for text classification models, in order to investigate the robustness of these models. It involves applying gradient-based perturbation on the sentence embeddings that are used as the features for the classifier, and learning a decoder for generation. We employ this method to a sentiment analysis model and verify its effectiveness in inducing incorrect predictions by the model. We also conduct quantitative and qualitative analysis on these examples and demonstrate that our approach can generate more natural adversaries. In addition, it can be used to successfully perform black-box attacks, which involves attacking other existing models whose parameters are not known. On a public sentiment analysis API, the proposed method introduces a 20% relative decrease in average accuracy and 74% relative increase in absolute error.

Chieh Hubert Lin, Chia-Che Chang, Yu-Sheng Chen et al.

Humans can only interact with part of the surrounding environment due to biological restrictions. Therefore, we learn to reason the spatial relationships across a series of observations to piece together the surrounding environment. Inspired by such behavior and the fact that machines also have computational constraints, we propose \underline{CO}nditional \underline{CO}ordinate GAN (COCO-GAN) of which the generator generates images by parts based on their spatial coordinates as the condition. On the other hand, the discriminator learns to justify realism across multiple assembled patches by global coherence, local appearance, and edge-crossing continuity. Despite the full images are never generated during training, we show that COCO-GAN can produce \textbf{state-of-the-art-quality} full images during inference. We further demonstrate a variety of novel applications enabled by teaching the network to be aware of coordinates. First, we perform extrapolation to the learned coordinate manifold and generate off-the-boundary patches. Combining with the originally generated full image, COCO-GAN can produce images that are larger than training samples, which we called "beyond-boundary generation". We then showcase panorama generation within a cylindrical coordinate system that inherently preserves horizontally cyclic topology. On the computation side, COCO-GAN has a built-in divide-and-conquer paradigm that reduces memory requisition during training and inference, provides high-parallelism, and can generate parts of images on-demand.

Hao-Yun Chen, Jhao-Hong Liang, Shih-Chieh Chang et al.

Adversarial robustness has emerged as an important topic in deep learning as carefully crafted attack samples can significantly disturb the performance of a model. Many recent methods have proposed to improve adversarial robustness by utilizing adversarial training or model distillation, which adds additional procedures to model training. In this paper, we propose a new training paradigm called Guided Complement Entropy (GCE) that is capable of achieving "adversarial defense for free," which involves no additional procedures in the process of improving adversarial robustness. In addition to maximizing model probabilities on the ground-truth class like cross-entropy, we neutralize its probabilities on the incorrect classes along with a "guided" term to balance between these two terms. We show in the experiments that our method achieves better model robustness with even better performance compared to the commonly used cross-entropy training objective. We also show that our method can be used orthogonal to adversarial training across well-known methods with noticeable robustness gain. To the best of our knowledge, our approach is the first one that improves model robustness without compromising performance.

Hao-Yun Chen, Pei-Hsin Wang, Chun-Hao Liu et al.

Learning with a primary objective, such as softmax cross entropy for classification and sequence generation, has been the norm for training deep neural networks for years. Although being a widely-adopted approach, using cross entropy as the primary objective exploits mostly the information from the ground-truth class for maximizing data likelihood, and largely ignores information from the complement (incorrect) classes. We argue that, in addition to the primary objective, training also using a complement objective that leverages information from the complement classes can be effective in improving model performance. This motivates us to study a new training paradigm that maximizes the likelihood of the groundtruth class while neutralizing the probabilities of the complement classes. We conduct extensive experiments on multiple tasks ranging from computer vision to natural language understanding. The experimental results confirm that, compared to the conventional training with just one primary objective, training also with the complement objective further improves the performance of the state-of-the-art models across all tasks. In addition to the accuracy improvement, we also show that models trained with both primary and complement objectives are more robust to single-step adversarial attacks.

Da-Cheng Juan, Chun-Ta Lu, Zhen Li et al.

Learning image representations to capture fine-grained semantics has been a challenging and important task enabling many applications such as image search and clustering. In this paper, we present Graph-Regularized Image Semantic Embedding (Graph-RISE), a large-scale neural graph learning framework that allows us to train embeddings to discriminate an unprecedented O(40M) ultra-fine-grained semantic labels. Graph-RISE outperforms state-of-the-art image embedding algorithms on several evaluation tasks, including image classification and triplet ranking. We provide case studies to demonstrate that, qualitatively, image retrieval based on Graph-RISE effectively captures semantics and, compared to the state-of-the-art, differentiates nuances at levels that are closer to human-perception.

An-Chieh Cheng, Chieh Hubert Lin, Da-Cheng Juan et al.

Conventional Neural Architecture Search (NAS) aims at finding a single architecture that achieves the best performance, which usually optimizes task related learning objectives such as accuracy. However, a single architecture may not be representative enough for the whole dataset with high diversity and variety. Intuitively, electing domain-expert architectures that are proficient in domain-specific features can further benefit architecture related objectives such as latency. In this paper, we propose InstaNAS---an instance-aware NAS framework---that employs a controller trained to search for a "distribution of architectures" instead of a single architecture; This allows the model to use sophisticated architectures for the difficult samples, which usually comes with large architecture related cost, and shallow architectures for those easy samples. During the inference phase, the controller assigns each of the unseen input samples with a domain expert architecture that can achieve high accuracy with customized inference costs. Experiments within a search space inspired by MobileNetV2 show InstaNAS can achieve up to 48.8% latency reduction without compromising accuracy on a series of datasets against MobileNetV2.

An-Chieh Cheng, Jin-Dong Dong, Chi-Hung Hsu et al.

Recent breakthroughs in Neural Architectural Search (NAS) have achieved state-of-the-art performance in many tasks such as image classification and language understanding. However, most existing works only optimize for model accuracy and largely ignore other important factors imposed by the underlying hardware and devices, such as latency and energy, when making inference. In this paper, we first introduce the problem of NAS and provide a survey on recent works. Then we deep dive into two recent advancements on extending NAS into multiple-objective frameworks: MONAS and DPP-Net. Both MONAS and DPP-Net are capable of optimizing accuracy and other objectives imposed by devices, searching for neural architectures that can be best deployed on a wide spectrum of devices: from embedded systems and mobile devices to workstations. Experimental results are poised to show that architectures found by MONAS and DPP-Net achieves Pareto optimality w.r.t the given objectives for various devices.

Chia-Che Chang, Chieh Hubert Lin, Che-Rung Lee et al.

Generative adversarial networks (GANs) often suffer from unpredictable mode-collapsing during training. We study the issue of mode collapse of Boundary Equilibrium Generative Adversarial Network (BEGAN), which is one of the state-of-the-art generative models. Despite its potential of generating high-quality images, we find that BEGAN tends to collapse at some modes after a period of training. We propose a new model, called \emph{BEGAN with a Constrained Space} (BEGAN-CS), which includes a latent-space constraint in the loss function. We show that BEGAN-CS can significantly improve training stability and suppress mode collapse without either increasing the model complexity or degrading the image quality. Further, we visualize the distribution of latent vectors to elucidate the effect of latent-space constraint. The experimental results show that our method has additional advantages of being able to train on small datasets and to generate images similar to a given real image yet with variations of designated attributes on-the-fly.

Chi-Hung Hsu, Shu-Huan Chang, Jhao-Hong Liang et al.

Recent studies on neural architecture search have shown that automatically designed neural networks perform as good as expert-crafted architectures. While most existing works aim at finding architectures that optimize the prediction accuracy, these architectures may have complexity and is therefore not suitable being deployed on certain computing environment (e.g., with limited power budgets). We propose MONAS, a framework for Multi-Objective Neural Architectural Search that employs reward functions considering both prediction accuracy and other important objectives (e.g., power consumption) when searching for neural network architectures. Experimental results showed that, compared to the state-ofthe-arts, models found by MONAS achieve comparable or better classification accuracy on computer vision applications, while satisfying the additional objectives such as peak power.

Jin-Dong Dong, An-Chieh Cheng, Da-Cheng Juan et al.

Recent breakthroughs in Neural Architectural Search (NAS) have achieved state-of-the-art performances in applications such as image classification and language modeling. However, these techniques typically ignore device-related objectives such as inference time, memory usage, and power consumption. Optimizing neural architecture for device-related objectives is immensely crucial for deploying deep networks on portable devices with limited computing resources. We propose DPP-Net: Device-aware Progressive Search for Pareto-optimal Neural Architectures, optimizing for both device-related (e.g., inference time and memory usage) and device-agnostic (e.g., accuracy and model size) objectives. DPP-Net employs a compact search space inspired by current state-of-the-art mobile CNNs, and further improves search efficiency by adopting progressive search (Liu et al. 2017). Experimental results on CIFAR-10 are poised to demonstrate the effectiveness of Pareto-optimal networks found by DPP-Net, for three different devices: (1) a workstation with Titan X GPU, (2) NVIDIA Jetson TX1 embedded system, and (3) mobile phone with ARM Cortex-A53. Compared to CondenseNet and NASNet (Mobile), DPP-Net achieves better performances: higher accuracy and shorter inference time on various devices. Additional experimental results show that models found by DPP-Net also achieve considerably-good performance on ImageNet as well.

Dimitrios Stamoulis, Ermao Cai, Da-Cheng Juan et al.

While selecting the hyper-parameters of Neural Networks (NNs) has been so far treated as an art, the emergence of more complex, deeper architectures poses increasingly more challenges to designers and Machine Learning (ML) practitioners, especially when power and memory constraints need to be considered. In this work, we propose HyperPower, a framework that enables efficient Bayesian optimization and random search in the context of power- and memory-constrained hyper-parameter optimization for NNs running on a given hardware platform. HyperPower is the first work (i) to show that power consumption can be used as a low-cost, a priori known constraint, and (ii) to propose predictive models for the power and memory of NNs executing on GPUs. Thanks to HyperPower, the number of function evaluations and the best test error achieved by a constraint-unaware method are reached up to 112.99x and 30.12x faster, respectively, while never considering invalid configurations. HyperPower significantly speeds up the hyper-parameter optimization, achieving up to 57.20x more function evaluations compared to constraint-unaware methods for a given time interval, effectively yielding significant accuracy improvements by up to 67.6%.

Ermao Cai, Da-Cheng Juan, Dimitrios Stamoulis et al.

"How much energy is consumed for an inference made by a convolutional neural network (CNN)?" With the increased popularity of CNNs deployed on the wide-spectrum of platforms (from mobile devices to workstations), the answer to this question has drawn significant attention. From lengthening battery life of mobile devices to reducing the energy bill of a datacenter, it is important to understand the energy efficiency of CNNs during serving for making an inference, before actually training the model. In this work, we propose NeuralPower: a layer-wise predictive framework based on sparse polynomial regression, for predicting the serving energy consumption of a CNN deployed on any GPU platform. Given the architecture of a CNN, NeuralPower provides an accurate prediction and breakdown for power and runtime across all layers in the whole network, helping machine learners quickly identify the power, runtime, or energy bottlenecks. We also propose the "energy-precision ratio" (EPR) metric to guide machine learners in selecting an energy-efficient CNN architecture that better trades off the energy consumption and prediction accuracy. The experimental results show that the prediction accuracy of the proposed NeuralPower outperforms the best published model to date, yielding an improvement in accuracy of up to 68.5%. We also assess the accuracy of predictions at the network level, by predicting the runtime, power, and energy of state-of-the-art CNN architectures, achieving an average accuracy of 88.24% in runtime, 88.34% in power, and 97.21% in energy. We comprehensively corroborate the effectiveness of NeuralPower as a powerful framework for machine learners by testing it on different GPU platforms and Deep Learning software tools.

You-Luen Lee, Da-Cheng Juan, Xuan-An Tseng et al.

When will a server fail catastrophically in an industrial datacenter? Is it possible to forecast these failures so preventive actions can be taken to increase the reliability of a datacenter? To answer these questions, we have studied what are probably the largest, publicly available datacenter traces, containing more than 104 million events from 12,500 machines. Among these samples, we observe and categorize three types of machine failures, all of which are catastrophic and may lead to information loss, or even worse, reliability degradation of a datacenter. We further propose a two-stage framework-DC-Prophet-based on One-Class Support Vector Machine and Random Forest. DC-Prophet extracts surprising patterns and accurately predicts the next failure of a machine. Experimental results show that DC-Prophet achieves an AUC of 0.93 in predicting the next machine failure, and a F3-score of 0.88 (out of 1). On average, DC-Prophet outperforms other classical machine learning methods by 39.45% in F3-score.

Da-Cheng Juan, Neil Shah, Mingyu Tang et al.

'Alice' is submitting one web search per five minutes, for three hours in a row - is it normal? How to detect abnormal search behaviors, among Alice and other users? Is there any distinct pattern in Alice's (or other users') search behavior? We studied what is probably the largest, publicly available, query log that contains more than 30 million queries from 0.6 million users. In this paper, we present a novel, user-and group-level framework, M3A: Model, MetaModel and Anomaly detection. For each user, we discover and explain a surprising, bi-modal pattern of the inter-arrival time (IAT) of landed queries (queries with user click-through). Specifically, the model Camel-Log is proposed to describe such an IAT distribution; we then notice the correlations among its parameters at the group level. Thus, we further propose the metamodel Meta-Click, to capture and explain the two-dimensional, heavy-tail distribution of the parameters. Combining Camel-Log and Meta-Click, the proposed M3A has the following strong points: (1) the accurate modeling of marginal IAT distribution, (2) quantitative interpretations, and (3) anomaly detection.