News on China's scientific and technological development.

latenlazy

Brigadier
Language is more than "text," it's also more about "context" and knowledge. To understand a piece of text, you often need to bring in the context and the necessary knowledge which are often implicit and unspoken. That's why large language models such as BERT are so large, because they're trained on huge amount of text and have so many parameters.

Yes, technically language or text is a one-dimensional sequence of symbols. Save for simple sentences, language is more than one dimension if you take into consideration of the context and implicit knowledge. One dimensional language can only express the simplest of ideas.
Right but that’s what I mean. If we think about what everyone is actually trying to do with AI recognition isn’t really the objective. Interpretation is. And interpretation is not fixed to a specific data medium.
 

supercat

Major
Is this a joke? The U.S. just finds out that the Huiwei ban also hurts themselves.

US eases Huawei curbs to counter China’s push on tech standards​


  • The Commerce Department’s Bureau of Industry and Security is issuing a new rule authorising the sharing of certain ‘low-level’ technologies and software
  • The Chinese tech giant was put on a blacklist in 2019, but the move also led US firms to limit their participation in standards-related activities
Please, Log in or Register to view URLs content!
 

AssassinsMace

Lieutenant General
Again the US thought no one can do anything without them hence why they thought they can control the standards. But now they've realized China can do without them meaning if China has its own standard and it'll spread across the world because like I've been saying everything the West does costs more therefore will be out of reach of most in the world, that means by default China's standard becomes mainstream and the US loses out because their standard is not compatible with the rest of the world. The US easing curbs is all about hoping to get China to go along before it's too late to follow the US.
 

BlackWindMnt

Captain
Registered Member
Again the US thought no one can do anything without them hence why they thought they can control the standards. But now they've realized China can do without them meaning if China has its own standard and it'll spread across the world because like I've been saying everything the West does costs more therefore will be out of reach of most in the world, that means by default China's standard becomes mainstream and the US loses out because their standard is not compatible with the rest of the world. The US easing curbs is all about hoping to get China to go along before it's too late to follow the US.
If im not mistaken Huawei has introduced a file system standard for mobile apps.
That recently got merged into android, wouldn't be surprised if Huawei will create application standards with open Harmony apps.
Kind of like flatpacks, AppImage etc you have right now for Linux.
 

xypher

Senior Member
Registered Member
Not an expert but this intuitively makes sense.

Text to me looks like 1D discrete dataset. There's only a given number of letters and there's no time dimension.

Audio looks like a 2D dataset that boils down to intensity vs time. From that you can derive frequency and all that but the raw input is simply intensity and time.

Video on the other hand is 6D data: each pixel can measure x-y positions, RGB values, time. Each new dimension added vastly increases the complexity.
It doesn't quite work like this. There are many complexities in each form of data, dimensionality alone is not the only challenge. For example, you can use raw RGB inputs for images and achieve good results but text requires extensive pre-processing to even work (tokenization, embeddings, etc.). Text and time series are also extremely tricky domains because they need context, both local and global - prior to emergence of transformers, it was very challenging to achieve good quality AND speed because recurrent neural networks were slow as heck and convolutional nets (even with dilations) handle global context poorly. Even now transformers are quite tricky because they require gigantic datasets to achieve decent performance, are prone to parameter bloat and vanilla transformers also have rather high complexity which makes them slow. This issue is intensified for mobile devices where CNNs are decently optimized but transformers still run poorly, so classic ML and even algorithmic approaches are still widely used for text. Another problem is with training such networks, there are multiple tricks on how to learn language models - e.g. classic "predict next token", BERT-style, GPT-styles etc.

So I would not say that one domain is easier or harder than the other one, there are complex issues and tricky subjects in all of them.
 

gadgetcool5

Senior Member
Registered Member
Is this a joke? The U.S. just finds out that the Huiwei ban also hurts themselves.

Please, Log in or Register to view URLs content!
Interesting. I wonder if China should ban Huawei from working with the US on standards. Maybe that would lock the US out of standards setting in this area. Or Huawei can just refuse to work with US standards setting unless the US govt makes concessions in other areas. Unlikely but...
 

tphuang

Lieutenant General
Staff member
Super Moderator
VIP Professional
Registered Member
Not 100% related, but kind of. This shows how Alibaba is spreading its wings in Pakistan and other countries in the region. This is also a great way to link up exporters with consumers in China and other places. It's not easy to make in road against Amazon, but Alibaba is trying to do that in many countries. It will be expanding to Europe next with Lazada I think. The fact that it has a group of exporters from places like Pakistan allow them to show case products that are not available on Amazon.
Please, Log in or Register to view URLs content!

Now that China has become the factory of the world, China is also trying to compete in the digital economy.
 

SanWenYu

Captain
Registered Member
A major breakthrough in conductive polymer by Chinese scientists.

Please, Log in or Register to view URLs content!

A solution-processed n-type conducting polymer with ultrahigh conductivity​


Abstract​

Conducting polymers (CPs) with high conductivity and solution-processability have gained great advances since the pioneering work on doped polyacetylene1-3, thus creating the new field of ‘organic synthetic metals’4. Various high-performance CPs have been realised, which enable the applications of multiple organic electronic devices5,6. Nevertheless, most CPs exhibit hole-dominant (p-type) transport behaviour7,8, while the development of n-type analogues lags far behind and few exhibits metallic state, typically limited by low doping efficiency and ambient instability. Here, we present a facilely synthesized highly conductive n-type polymer poly(benzodifurandione) (PBFDO). The reaction combines oxidative polymerisation and in situ reductive n-doping, dramatically increasing the doping efficiency, and doping level of almost 0.9 charge per repeating unit can be achieved. The resultant polymer exhibits a breakthrough conductivity over 2000 S cm−1 with excellent stability and an unexpected solution-processability without extra side chains or surfactants. Additionally, detailed investigations on PBFDO reveal coherent charge transport properties and existence of metallic state. The benchmark performances in electrochemical transistors and thermoelectric generators are further demonstrated, thus paving the way for application of the n-type CPs in organic electronics.

Please, Log in or Register to view URLs content!

导电聚合物已经在太阳能电池、传感器和一些显示技术中得到了十分广泛的应用。自聚乙炔作为第一种被发现的导电聚合物被发现以来,导电聚合物的导电性能已经可以达到1000 S cm-1以上。由于绝大多数导电聚合物是一种具有共轭结构的富电子聚合物,因此这类聚合物的更加容易实现的是被氧化形成空穴,并以空穴作为导电的载流子,这类材料即使p型导电高分子。出于共轭高分子富电子的特性,导电高分子在于掺杂剂发生还原反应的过程中通常受到低掺杂效率和环境不稳定性的限制,这就导致了载流子为电子的导电高分子(n型导电高分子)的发展远远落后于p型导电高分子。

为了实现高导电性的n型导电聚合物,应同时获得高效的电子传递速率和高载流子浓度。首先,需要设计具有扩展共轭框架的大型和刚性骨架,使得极化子易于离域,同时让载流子在聚合物链上易于传输;其次,需要选择适合的掺杂剂和聚合物充分反应,增加载流子的浓度。然而,刚性骨架十分难以溶解,需要使用引入侧链或表面活性剂的方式进行功能化,以确保这类聚合物的溶液性和可加工性,而这种处理方式将对电导率产生不利影响。另一个问题是大多数n型导电聚合物的掺杂效率相当低(通常在10%左右),通常的解决方式是降低导电聚合物最低未占有轨道(LUMO)的能量或设计空气稳定的n型掺杂剂。而这就需要设计十分复杂的分子结构,导致这类导电性能得到提高的n型导电高分子难以被应用。

近期,华南理工大学黄飞教授、曹镛院士、马於光院士和北京大学裴坚教授、南方科技大学郭旭岗教授等合作在这一领域取得重大突破。他们提出了一种易于合成的高导电性n型聚合物,聚苯二呋喃二酮(PBFDO)。该聚合物单体的聚合反应结合了氧化聚合和原位还原n掺杂,这种方式显著提高了掺杂效率,实现每个重复单元可以达到近0.9个电荷的掺杂水平。所得到的聚合物在具有创纪录的电导率,可达到2000 S cm-1以上,并且具有优异的稳定性,在没有额外的侧链或表面活性剂的情况下具有良好的溶液处理性。此外,该团队还发现了PBFDO具有相干电荷输运特性和金属态的存在,进一步证明了该材料具备电化学晶体管和热电发电机所需的基准性能,从而为这种n型导电高分子在有机电子学中的应用铺平了道路。该工作以题为“A solution-processed n-type conducting polymer with ultrahigh conductivity”的文章发表于Nature上。
 
Top