Everyday AI

人工智慧，形影相隨

MediaTek

MediaTek is the world’s 4th largest global fabless semiconductor company. We are a market leader in developing innovative systems-on-chip (SoC) for mobile devices, home entertainment, connectivity and IoT products. Ultimately, we power more than 2 billion devices a year – that’s in 20 percent of homes and nearly 1 of every 3 mobile phones globally.

MediaTek Research, the AI research arm of MediaTek, brings the most advanced developments in AI to everyday devices. In addition to designing embedded AI systems, MediaTek Research contributes to the fundamental research of AI-based algorithmic solutions.

News and Events

2022-06-14

MediaTek Announces Breakthrough in Artificial Intelligence and Chip Design
2021-11-16

MediaTek Research opened a brand new office in National Taiwan University
2022-10-25
AI breaks into IC design! Deep learning algorithm is showing its power
2022-05-05

MediaTek Announces Breakthrough in Artificial Intelligence and Chip Design
2021-11-16

MediaTek Research opened a brand new office in National Taiwan University
Read more
2022-10-25
AI breaks into IC design! Deep learning algorithm is showing its power
2022-05-05

MediaTek Announces Breakthrough in Artificial Intelligence and Chip Design
2021-11-16

MediaTek Research opened a brand new office in National Taiwan University
2021-11-12

MediaTek Research has 6 papers accepted at NeurIPS 2021 conference and workshops
Read more

Research

Papers

A Learning-Based Algorithm for Early Floorplan with Flexible Blocks

This paper presents a learning-based algorithm using graph neural network (GNN) and deconvolution network to predict the placement of the locations and the aspect ratios for the design blocks with flexible rectangles. With several hours of training on 4 GPUs, the proposed method, targeting at minimizing the cost of wirelength, can generate the placements in early stage of floorplan which is superior to that from the manual placements which requires several days’ efforts for physical design experts.

Gradient Descent: Robustness to Adversarial Corruption

Optimization using gradient descent (GD) is a ubiquitous practice in various machine learning problems including training large neural networks. Noise-free GD and stochastic GD--corrupted by random noise--have been extensively studied in the literature, but less attention has been paid to an adversarial setting, that is subject to adversarial corruptions in the gradient values. In this work, we analyze the performance of GD under a proposed general adversarial framework. For the class of functions satisfying the Polyak-Łojasiewicz condition, we derive finite time bounds on a minimax optimization error. Based on this bound, we provide a guideline on the choice of learning rate sequence with theoretical guarantees on the robustness of GD against adversarial corruption.

Near-Optimal Collaborative Learning in Bandits

This paper introduces a general multi-agent bandit model in which each agent is facing a finite set of arms and may communicate with other agents through a central controller in order to identify -in pure exploration- or play -in regret minimization- its optimal arm. The twist is that the optimal arm for each agent is the arm with largest expected mixed reward, where the mixed reward of an arm is a weighted sum of the rewards of this arm for all agents. This makes communication between agents often necessary. This general setting allows to recover and extend several recent models for collaborative bandit learning, including the recently proposed federated learning with personalization [Shi et al., 2021]. In this paper, we provide new lower bounds on the sample complexity of pure exploration and on the regret. We then propose a near-optimal algorithm for pure exploration. This algorithm is based on phased elimination with two novel ingredients: a data-dependent sampling scheme within each phase, aimed at matching a relaxation of the lower bound.
Improved Convergence Rates for Sparse Approximation Methods in Kernel-Based Learning

Kernel-based models such as kernel ridge regression and Gaussian processes are ubiquitous in machine learning applications for regression and optimization. It is well known that a major downside for kernel-based models is the high computational cost; given a dataset of n samples, the cost grows as O(n3). Existing sparse approximation methods can yield a significant reduction in the computational cost, effectively reducing the actual cost down to as low as O(n) in certain cases. Despite this remarkable empirical success, significant gaps remain in the existing results for the analytical bounds on the error due to approximation. In this work, we provide novel confidence intervals for the Nyström method and the sparse variational Gaussian process approximation method, which we establish using novel interpretations of the approximate (surrogate) posterior variance of the models. Our confidence intervals lead to improved performance bounds in both regression and optimization problems.
Regret Bounds for Noise-Free Kernel-Based Bandits

Kernel-based bandit is an extensively studied black-box optimization problem, in which the objective function is assumed to live in a known reproducing kernel Hilbert space. While nearly optimal regret bounds (up to logarithmic factors) are established in the noisy setting, surprisingly, less is known about the noise-free setting (when the exact values of the underlying function is accessible without observation noise). We discuss several upper bounds on regret; none of which seem order optimal, and provide a conjecture on the order optimal regret bound.
LPI: Learned Positional Invariances for Transfer of Task Structure and Zero-shot Planning

Real-world tasks often include interactions with the environment where our actions can drastically change the available or desirable long-term outcomes. One formulation of this in the reinforcement learning setting is in terms of nonMarkovian rewards. Here the reward function, and thus the available rewards, are themselves history-dependent, and dynamically change given the agent-environment interactions. An important challenge for navigating such environments is to be able to capture the structure of this dynamic reward function, in a way that is interpretable and allows for optimal planning. This structure, in conjunction with the particular task setting at hand, then determines the optimal order in which actions should be executed, or subtasks completed. Planning methods face the challenge of combinatorial explosion if all such orderings need to be evaluated, however, learning invariances inherent in the task structure can alleviate this pressure. Here we propose a solution to this problem by allowing the planning method to recognise task segments where temporal ordering is irrelevant for predicting reward outcomes downstream. To facilitate this, our agent simultaneously learns to segment a task and predict the changing reward function resulting from its actions, while also learning about the permutation invariances in the its history that are relevant for this prediction. This dual approach can allow zero-shot or few-shot generalisation for complex, dynamic reinforcement learning tasks
Adaptive Erasure of Spurious Sequences in Sensory Cortical Circuits

Sequential activity reflecting previously experienced temporal sequences is considered a hallmark of learning across cortical areas. However, it is unknown how cortical circuits avoid the converse problem: producing spurious sequences that are not reflecting sequences in their inputs. We develop methods to quantify and study sequentiality in neural responses. We show that recurrent circuit responses generally include spurious sequences, which are specifically prevented in circuits that obey two widely known features of cortical microcircuit organization: Dale’s law and Hebbian connectivity. In particular, spike-timing-dependent plasticity in excitation-inhibition networks leads to an adaptive erasure of spurious sequences. We tested our theory in multielectrode recordings from the visual cortex of awake ferrets. Although responses to natural stimuli were largely non-sequential, responses to artificial stimuli initially included spurious sequences, which diminished over extended exposure. These results reveal an unexpected role for Hebbian experience-dependent plasticity and Dale’s law in sensory cortical circuits.
Flexible Multiple-Objective Reinforcement Learning for Chip Placement

Recently, successful applications of reinforcement learning to chip placement have emerged. Pretrained models are necessary to improve efficiency and effectiveness. Currently, the weights of objective metrics (e.g., wirelength, congestion, and timing) are fixed during pretraining. However, fixed-weighed models cannot generate the diversity of placements required for engineers to accommodate changing requirements as they arise. This paper proposes flexible multiple-objective reinforcement learning (MORL) to support objective functions with inference-time variable weights using just a single pretrained model. Our macro placement results show that MORL can generate the Pareto frontier of multiple objectives effectively.

A Learning-Based Algorithm for Early Floorplan with Flexible Blocks

This paper presents a learning-based algorithm using graph neural network (GNN) and deconvolution network to predict the placement of the locations and the aspect ratios for the design blocks with flexible rectangles. With several hours of training on 4 GPUs, the proposed method, targeting at minimizing the cost of wirelength, can generate the placements in early stage of floorplan which is superior to that from the manual placements which requires several days’ efforts for physical design experts.

Regret Bounds for Noise-Free Kernel-Based Bandits

Kernel-based bandit is an extensively studied black-box optimization problem, in which the objective function is assumed to live in a known reproducing kernel Hilbert space. While nearly optimal regret bounds (up to logarithmic factors) are established in the noisy setting, surprisingly, less is known about the noise-free setting (when the exact values of the underlying function is accessible without observation noise). We discuss several upper bounds on regret; none of which seem order optimal, and provide a conjecture on the order optimal regret bound.
Gradient Descent: Robustness to Adversarial Corruption

Optimization using gradient descent (GD) is a ubiquitous practice in various machine learning problems including training large neural networks. Noise-free GD and stochastic GD--corrupted by random noise--have been extensively studied in the literature, but less attention has been paid to an adversarial setting, that is subject to adversarial corruptions in the gradient values. In this work, we analyze the performance of GD under a proposed general adversarial framework. For the class of functions satisfying the Polyak-Łojasiewicz condition, we derive finite time bounds on a minimax optimization error. Based on this bound, we provide a guideline on the choice of learning rate sequence with theoretical guarantees on the robustness of GD against adversarial corruption.

LPI: Learned Positional Invariances for Transfer of Task Structure and Zero-shot Planning

Real-world tasks often include interactions with the environment where our actions can drastically change the available or desirable long-term outcomes. One formulation of this in the reinforcement learning setting is in terms of nonMarkovian rewards. Here the reward function, and thus the available rewards, are themselves history-dependent, and dynamically change given the agent-environment interactions. An important challenge for navigating such environments is to be able to capture the structure of this dynamic reward function, in a way that is interpretable and allows for optimal planning. This structure, in conjunction with the particular task setting at hand, then determines the optimal order in which actions should be executed, or subtasks completed. Planning methods face the challenge of combinatorial explosion if all such orderings need to be evaluated, however, learning invariances inherent in the task structure can alleviate this pressure. Here we propose a solution to this problem by allowing the planning method to recognise task segments where temporal ordering is irrelevant for predicting reward outcomes downstream. To facilitate this, our agent simultaneously learns to segment a task and predict the changing reward function resulting from its actions, while also learning about the permutation invariances in the its history that are relevant for this prediction. This dual approach can allow zero-shot or few-shot generalisation for complex, dynamic reinforcement learning tasks.
Near-Optimal Collaborative Learning in Bandits

This paper introduces a general multi-agent bandit model in which each agent is facing a finite set of arms and may communicate with other agents through a central controller in order to identify -in pure exploration- or play -in regret minimization- its optimal arm. The twist is that the optimal arm for each agent is the arm with largest expected mixed reward, where the mixed reward of an arm is a weighted sum of the rewards of this arm for all agents. This makes communication between agents often necessary. This general setting allows to recover and extend several recent models for collaborative bandit learning, including the recently proposed federated learning with personalization [Shi et al., 2021]. In this paper, we provide new lower bounds on the sample complexity of pure exploration and on the regret. We then propose a near-optimal algorithm for pure exploration. This algorithm is based on phased elimination with two novel ingredients: a data-dependent sampling scheme within each phase, aimed at matching a relaxation of the lower bound.
Adaptive Erasure of Spurious Sequences in Sensory Cortical Circuits

Sequential activity reflecting previously experienced temporal sequences is considered a hallmark of learning across cortical areas. However, it is unknown how cortical circuits avoid the converse problem: producing spurious sequences that are not reflecting sequences in their inputs. We develop methods to quantify and study sequentiality in neural responses. We show that recurrent circuit responses generally include spurious sequences, which are specifically prevented in circuits that obey two widely known features of cortical microcircuit organization: Dale’s law and Hebbian connectivity. In particular, spike-timing-dependent plasticity in excitation-inhibition networks leads to an adaptive erasure of spurious sequences. We tested our theory in multielectrode recordings from the visual cortex of awake ferrets. Although responses to natural stimuli were largely non-sequential, responses to artificial stimuli initially included spurious sequences, which diminished over extended exposure. These results reveal an unexpected role for Hebbian experience-dependent plasticity and Dale’s law in sensory cortical circuits.
Improved Convergence Rates for Sparse Approximation Methods in Kernel-Based Learning

Kernel-based models, such as kernel ridge regression and Gaussian processes, are widely used in machine learning for regression and optimization. It is well known that a major disadvantage of the core model is high computational cost; given a dataset consisting of n samples, the computational cost grows in O(n3). Existing sparse approximation methods can reduce the computational cost, in some practical cases, down to O(n). Despite this surprising empirical result, the error value of the approximation method, which limits the analytical precision, still leaves a significant gap in the existing results.

Flexible Multiple-Objective Reinforcement Learning for Chip Placement

Recently, successful applications of reinforcement learning to chip placement have emerged. Pretrained models are necessary to improve efficiency and effectiveness. Currently, the weights of objective metrics (e.g., wirelength, congestion, and timing) are fixed during pretraining. However, fixed-weighed models cannot generate the diversity of placements required for engineers to accommodate changing requirements as they arise. This paper proposes flexible multiple-objective reinforcement learning (MORL) to support objective functions with inference-time variable weights using just a single pretrained model. Our macro placement results show that MORL can generate the Pareto frontier of multiple objectives effectively.

Read more

Talks

Improved Convergence Rates for Sparse Approximation Methods in Kernel-Based Learning

Kernel-based models such as kernel ridge regression and Gaussian processes are ubiquitous in machine learning applications for regression and optimization. It is well known that a serious downside for kernel-based models is the high computational cost; given a dataset of n samples, the cost grows as O(n^3).

An Introduction to the Mean-Field Approach for Neural Networks

This talk is the first part of a short series where we will spend some time trying to learn the theory and usage of what is probably the most beautiful applications form dynamical systems to computational science, namely the mean-field approach to Neural Network SGD dynamics.

Polyak-type step sizes for mirror descent methods

In this talk, we will first introduce the standard Polyak step size, which recently is an attractive learning rate for stochastic gradient descent (SGD) when training over-parametrized models. We then propose two variants of the standard Polyak step size for mirror descent and prove their convergences for minimizing convex locally Lipschitz function.

Optimal Order Simple Regret for Gaussian Process Bandits

In this talk, Mark will give us a talk on the Gaussian process bandit problem. In this work, an algorithm called Maximum Variance Reduction is proposed, which can guarantee that the simple regret of Gaussian process bandit will be converged to zero when number of trials approach infinity.

Read more

Blog

MediaTek Announces Breakthrough in Artificial Intelligence and Chip Design

This July, MediaTek will be presenting joint research with College of Electrical Engineering & Computer Science, National Taiwan University (NTU EECS) and Maxeda Technology at the ACM/IEEE Design Automation Conference (DAC), the most influential and longest-running conference on Electronic Design Automation (EDA) in the world. As an added distinction, this year’s DAC has selected the said research as one of its inaugural Publicity Papers. MediaTek collaborates with NTU EECS and Maxeda Technology to introduce a new AI-powered chip design algorithm at the ACM/IEEE Design Automation Conference The multiple-objective reinforcement learning algorithm proposed in the research can simultaneously optimize conflicting objectives in chip design in terms of power consumption, efficiency, chip area, yield, etc. It opens the door to significantly reduce development costs, cut down development time, and improve chip performance. The algorithm has already been applied to the creation of the MediaTek Dimensity mobile chip series and will be used in other product lines as well. “MediaTek strives to stay ahead of the curve in technology,” said SR Tsai, Corporate Senior Vice President of MediaTek’s Central Design Group. “To develop such cutting-edge technology, we employ artificial intelligence in certain aspects of the chip design process to make up for a…

Non-reversible Gaussian Processes

In Machine Learning, it is often convenient to use functions that are not fixed but only known with some probability. For example, let say we know the weather in Cambridge for the last few years and we would like to predict what will be the weather tomorrow (e.g. the temperature or wind speed, a function of time $f(t)$). Or, we know today’s weather in Cambridge and London, and we would like to know what is happening along the M11 motorway between the two cities. Instead of giving one predicted value, we would like a probability distribution for each time (or space point). Since time is continuous, we are looking for a distribution over functions $f(t)$. Gaussian Processes (GP) represent a simple and powerful example of distribution over functions. They are widely used for modeling and prediction (Rasmussen and Williams 2006), a few examples of GPs are shown in Figure 1. Perhaps the most important feature of a GP is the correlation function $k(t,t’)$, which determines the information we get about $f(t’)$ when measuring the true value of $f(t)$. The shape of this correlation function is the most important choice to make when we use a GP, it determines what kind…

TreeMAML

Many NLP models rely on training in one high-resource language, and they cannot be directly used to make predictions  for other languages at inference time. Most of the languages of the world are under-resourced and rely on Machine Translation (MT) to English to make use of Language Models. However, having an MT system in every direction is costly and  not the best solution for every NLP task. We propose the use of meta-learning to solve this issue. Our algorithm, TreeMAML, extends a meta-learning model, MAML [1], by exploiting hierarchical languages relationships. MAML adapts the model to each task with a few gradient steps. In our method, TreeMAML, this adaptation follows the hierarchical tree structure:In each step down the tree, gradients are pooled across language clusters: Algorithm 1.Algorithm 2 is a non-binary modification of the OTD clustering [2],  that generates the language tree without previous knowledge of the structure, allowing us to use implicit relationships between the languages.In our Experiments we adapt a high-resource language model, Multi-BERT [3], to a Few-Shot NLI task with these steps:We use the XNLI data set [4]. XNLI corpus is a crowd-sourced collection of pairs for the MultiNLI corpus with 10 different genres in 15 languages. The pairs are…

Non-reversible Gaussian Processes

In Machine Learning, it is often convenient to use functions that are not fixed but only known with some probability. For example, let say we…

TreeMAML

Many NLP models rely on training in one high-resource language, and they cannot be directly used to make predictions  for other languages at inference time. Most…

Read more

About Us

We at MediaTek Research are a dedicated research unit of the global MediaTek Group. Our mission is to explore and initiate emerging approaches in AI and ML from both fundamental and applied perspectives. We have two research centers, one in Cambridge (UK), and one in National Taiwan University. We cooperate closely with worldwide academic community.

Our team is made of researchers with a diverse background in computer science, engineering, mathematics and physics.

Previous
Next

Contact Us

Interested in knowing more about MediaTek Research? Please feel free to contact us:

 

Our Email Address: info@mtkresearch.com