Core Viewpoint - The company TOLAN Food is showcasing its dual AI optical sorting platforms, Spectrim and InVision², equipped with the LUCAi™ artificial intelligence engine, at the 2025 China International Fruit Expo and Asia International Fruit and Vegetable Exhibition, marking the first simultaneous exhibition of these technologies in China, aimed at driving innovation in the fruit sorting industry [1][2]. Group 1: Technology and Innovation - TOLAN has been a leader in the sorting field for over 50 years and has focused on AI and food sorting technology since 2010, developing the LUCAi™ AI engine that utilizes deep learning to accurately identify various fruit defects [2][3]. - The dual AI optical sorting platforms leverage continuously iterated algorithms of the LUCAi™ engine to enhance the sorting process, overcoming challenges in identifying complex fruit defects that traditional machine learning struggles with [2][3]. - The LUCAi™ engine can process up to 40,000 images per second, significantly improving defect recognition capabilities compared to traditional sorting systems, thus providing precise quality metrics for fruits from harvest to market [3]. Group 2: Market Strategy and Localization - TOLAN is committed to deepening its localization strategy in the Chinese market, having established regional headquarters, manufacturing bases, supply chain centers, and technical testing centers since entering China in 2010 [5]. - The company’s AI sorting systems can utilize localized data to upgrade algorithms and effectively identify regional fruit defects, ensuring quality that meets local demands [5]. - As the largest producer and consumer of fruits globally, China represents a strategic market for TOLAN, which aims to leverage its dual AI solutions to help clients capture high-value markets and promote the standardization of AI sorting in the Chinese fruit industry [5].

Core Viewpoint - The article discusses the evolution and potential of AI in the music industry, highlighting the journey of a company focused on AI music generation and the belief in democratizing music creation for everyone [6][10]. Group 1: Historical Context of AI in Music - The first electronic speech synthesizer, Voder, was built in 1938, marking the initial connection between AI and audio [7]. - In 1957, the first computer-generated music piece, "Illiac Suite," was created, but progress in AI music was slow for decades [7]. - The introduction of Google's Magenta project in 2016 showcased the capabilities of AI in music generation, leading to significant advancements in the field [8]. Group 2: Personal Journey and Company Development - The CEO of the company, who has a background in AI algorithms, experienced a pivotal moment in 2016 when he successfully separated vocals from accompaniment using deep learning techniques [8][9]. - The company was founded in 2021, aiming to create a platform where everyone can compose music, similar to how short videos democratized content creation [10][11]. - The CEO believes that music creation can also achieve equality, allowing diverse voices and stories to be expressed through music [10][11]. Group 3: Technological Innovations and Challenges - The emergence of large models based on the Transformer architecture in 2021 led to significant advancements in AI music generation, culminating in the launch of a product referred to as the "ChatGPT of music" [9][10]. - The company is focused on rapid product iterations, aiming to enhance user engagement and creativity through innovative features [39][48]. - The challenge lies in stimulating user creativity and finding effective ways to shorten the music creation process [45][46]. Group 4: Business Model and Market Positioning - The company plans to offer a freemium model, allowing users to create a limited number of songs for free, with monetization options based on song popularity [52][53]. - A significant effort has been made to build a comprehensive music data labeling database, which serves as a competitive advantage in the AI music space [54]. - The company aims to differentiate itself from competitors by focusing on user-generated content and providing a platform for music creation that emphasizes user ownership of their work [55][61].

Core Insights - The article highlights the prominence of AI as the hottest research direction globally, with Yoshua Bengio being the most cited scientist ever, accumulating a total citation count of 973,655 and 698,008 citations in the last five years [1][3]. Group 1: Citation Rankings - The AD Scientific Index ranks 2,626,749 scientists from 221 countries and 24,576 institutions based on total citation counts and recent citation indices [3]. - Yoshua Bengio's work on Generative Adversarial Networks (GANs) has surpassed 100,000 citations, outpacing his co-authored paper "Deep Learning," which also exceeds 100,000 citations [3][4]. - Geoffrey Hinton, a pioneer in AI, ranks second with over 950,000 total citations and more than 570,000 citations in the last five years [4][5]. Group 2: Notable Papers and Their Impact - The paper "AlexNet," co-authored by Hinton, Krizhevsky, and Sutskever, has received over 180,000 citations, marking a significant breakthrough in deep learning for computer vision [5][6]. - Kaiming He’s paper "Deep Residual Learning for Image Recognition" has over 290,000 citations, establishing ResNet as a foundational model in modern deep learning [10][11]. - The article notes that ResNet is recognized as the most cited paper of the 21st century, with citation counts ranging from 103,756 to 254,074 across various databases [11]. Group 3: Broader Implications - The high citation counts of these influential papers indicate their lasting impact on the academic community and their role in shaping future research directions in AI and related fields [17].

Core Viewpoint - Yoshua Bengio is recognized as the most cited living scientist across all disciplines, not just in computer science, highlighting his significant impact on deep learning and artificial intelligence [4][19]. Group 1: Background and Contributions - Yoshua Bengio, born in 1964 in Paris, is a prominent figure in deep learning, having co-founded the field alongside Geoffrey Hinton and Yann LeCun [8][11]. - His early academic journey included a PhD under Hinton at McGill University, where he shifted focus from classical statistical models to neural networks [10][12]. - Bengio's major contributions include the development of probabilistic modeling, high-dimensional word embeddings, attention mechanisms, and generative adversarial networks (GANs) [13][16]. Group 2: Key Publications - Bengio's influential papers include "A Neural Probabilistic Language Model" (2000), which addressed the "curse of dimensionality" in language modeling, laying the groundwork for modern language models [14]. - The paper "Generative Adversarial Nets" (2014), co-authored with Ian Goodfellow, is his most cited work, with over 100,904 citations [17]. - The 2015 paper "Deep Learning," co-authored with Hinton and LeCun, is considered a foundational text in the field, summarizing deep learning's evolution and theoretical underpinnings [16][17]. Group 3: Recent Developments - In June 2023, Bengio announced the establishment of a non-profit organization, LawZero, aimed at developing the next generation of AI systems, with an initial funding of $30 million [19][20]. - LawZero focuses on understanding the learning world rather than action-oriented AI, aiming to provide verifiable answers to enhance scientific discovery and address AI risks [20]. Group 4: Citation Rankings - Bengio currently leads in citation counts among living scientists, with his closest competitor being Geoffrey Hinton, who has nearly 940,000 citations [21]. - The AD Scientific Index ranks researchers based on various metrics, including total citations, reflecting the prominence of AI and medical research in current academic discourse [23][26].

Core Viewpoint - Horizon has launched its most significant upgrade of the high-performance urban auxiliary driving product HSD, which is recognized as the "most Tesla-like FSD system" in China, enhancing safety, efficiency, and comfort in driving [3][5]. Group 1: Product Development and Features - HSD is built on the Journey 6P platform, achieving ultra-low latency from "photon input to trajectory output" and introducing reinforcement learning for maximizing model potential [8]. - The product covers various driving scenarios, including urban areas, highways, rural roads, and parking lots, and can handle complex driving tasks without relying on memory mapping [3][5]. - Horizon has established cooperation intentions with nearly 10 global automotive brands, with the first mass production set to be launched on Chery's Exeed Star Era E05 [3][5]. Group 2: Market Trends and Growth - The high-level auxiliary driving technology is transitioning from validation to large-scale adoption, with 902,200 new cars equipped with urban NOA delivered in China in the first five months of the year, marking a 152.5% year-on-year increase [5]. - Horizon has achieved over 8 million sets of front-mounted mass production shipments and over 200 mass-produced vehicle models as of the end of Q1 this year [16]. Group 3: Strategic Vision and Competitive Edge - The founder and CEO of Horizon emphasizes the importance of continuous technological iteration and the launch of high-performance products to maintain market leadership [16]. - Horizon's strategy of "soft and hard integration" has allowed it to achieve over 1000 times improvement in computing performance over the past decade, supporting the mass production of over 10 computing solutions [19][21]. - The company aims to become a leading player in the high-level auxiliary driving market, leveraging its unique position as a "full-stack soft and hard technology enterprise" with extensive practical experience [23].

Core Viewpoint - The recent advancements in large model technology have renewed interest in AI-specific chips, particularly Google's TPU, which has evolved significantly since its deployment in 2015, now reaching its 7th generation [1][9]. Group 1: TPU Overview - TPU is a specialized chip designed by Google to enhance the speed of machine learning model inference and training, focusing on executing mathematical operations efficiently [9]. - The architecture of TPU allows it to perform matrix multiplication efficiently, which constitutes a significant portion of computations in deep learning models [14][31]. Group 2: TinyTPU Project - The TinyTPU project was initiated by engineers from Western University in Canada to create an open-source ML inference and training chip, motivated by the lack of a complete open-source codebase for such accelerators [5][7]. - The project emphasizes a hands-on approach to learning hardware design and deep learning principles, avoiding reliance on AI tools for coding [6]. Group 3: Hardware Design Insights - The project team established a design philosophy of exploring unconventional ideas before consulting external resources, leading to the re-invention of many key mechanisms used in TPU [6]. - The hardware design process involves understanding clock cycles, using Verilog for hardware description, and implementing a systolic array architecture for efficient matrix multiplication [10][12][26]. Group 4: Training and Inference Mechanisms - The TinyTPU architecture allows for continuous inference by utilizing a double buffering mechanism, which enables the loading of new weights while processing current computations [61][64]. - The training process leverages the same architecture as inference, with additional modules for gradient calculation and weight updates, allowing for efficient training of neural networks [71][118]. Group 5: Control and Instruction Set - The control unit of TinyTPU employs a custom instruction set architecture (ISA) to manage control signals and data flow, enhancing the efficiency of operations [68][117]. - The ISA has evolved to include 94 bits, ensuring that all necessary control flags and data fields are accounted for without compromising performance [117].

Core Insights - Taobao is accelerating the implementation of AI technology in consumer scenarios with a new feature called "AI Universal Search" currently in gray testing [3] - This innovative search function represents a significant transformation in the e-commerce search model, moving away from traditional keyword matching to a conversational interaction approach [3][4] - The system can understand user queries in natural language and generate a comprehensive "answer report" that includes product links, review videos, and purchasing guides [3][4] Feature Details - "AI Universal Search" allows users to ask questions in everyday language, such as "What are some simple style dresses suitable for new employees?" or "Recommendations for practical gifts under 500 yuan for my father?" [3] - The system breaks down key dimensions like cost-performance ratio, budget range, and battery life when users input queries like "How to choose a phone," providing a layered product recommendation scheme [4] - Users receive tailored "avoid pitfalls" reminders and pairing suggestions, with the system guiding them to refine their queries if they are not satisfied with the results [4] Technical Capabilities - The feature relies on Alibaba Cloud's Tongyi large model technology, combined with Taobao's vast product data and user behavior insights, enabling strong semantic understanding and content generation capabilities [4] - The system dynamically organizes information based on user needs, providing personalized recommendations, such as suggesting air conditioners suitable for small apartments along with installation tips and user reviews [4] - "AI Universal Search" also incorporates a "shopping preference" function using collaborative filtering algorithms, allowing the AI to understand user tastes and preferences, achieving a level of personalization previously unattainable by other platforms [4] Additional Information - It remains unclear whether "AI Universal Search" utilizes other models like DeepSeek in addition to Tongyi Qianwen, and whether the search data is based on product details or user-generated content [5]

Section 1: Investment Rating of the Reported Industry - The provided content does not mention the industry investment rating [1][2] Section 2: Core Views of the Report - Deep learning may be used for convertible bond pricing. Based on the Universal Approximation Theorem (UAT), if there is a reasonable analytical solution for convertible bond pricing, a neural network model can fit the result [2][5] - A Multilayer Perceptron (MLP) model is designed. It uses 11 factors, including core factors, convertible bond-specific factors, and market performance factors, to nonlinearly fit the pricing characteristics of convertible bonds [2][5] - The MLP model has good convergence and excellent extrapolation generalization ability. It can strongly explain out-of-sample data from 2024 to Q1 2025 [2][8] - After developing multiple models, including the MLP, MC, and traditional BS models, they can assist in investment activities in various scenarios such as new bond pricing, market interpretation, and clause pricing [2][13] - The neural network model indicates that the current market pricing of convertible bonds is overestimated, but not as much as expected. Convertible bond valuations are high but may still rise further [2][13] - In new bond pricing, the MLP and MC models form a "high-low combination." The MC model is better at pricing large-scale, high-rated convertible bonds, while the MLP is more effective for regular convertible bond listings [2][16] - The models also work well for pricing convertible bond downward revisions [2][19] Section 3: Summary by Relevant Catalog 1. Deep Learning Pricing Model's Concept and Design - The MLP model is based on the idea that if there is an analytical solution for convertible bond pricing, a neural network can fit it. It uses 11 factors for pricing [5] - The model has good convergence and generalization ability. After training with data from 2022 - 2023 and cleaning the dataset, it can effectively explain out-of-sample data from 2024 to Q1 2025 [8] - Compared with the BS and MC models, the MLP model has better pricing results for the overall market and individual convertible bonds. It has faster computation speed than the MC model and is more suitable for real - world scenarios than the BS model. However, it has limitations such as being a "black box" and requiring a large amount of historical data [10][11] 2. Convertible Bond Quantitative Pricing 2.0 - What Are the Model's Applications? - With multiple models (MLP, MC, and BS), they can assist in investment activities in various scenarios [13] - At the overall market pricing level, the neural network model shows that the current market pricing is overestimated, but not significantly. Convertible bond valuations are high but may still increase [13] - In new bond pricing, the MLP and MC models complement each other. The MC model is better for large - scale, high - rated convertible bonds, and the MLP is better for regular convertible bonds. Over 50% of convertible bond listing prices fall within the range defined by the two models, and over 80% are captured after the pricing repair in November 2024 [16] - For downward revision pricing, the MLP model can predict prices when the convertible bond is revised to the trigger threshold and to the lowest level. Most convertible bond prices on the second trading day after a downward revision proposal fall within or near this predicted range [19]

Group 1: Core Insights - The first humanoid robot sports competition was held in Beijing, featuring 280 teams and over 500 humanoid robots from 16 countries, marking a significant step for humanoid robots from laboratory to real-world applications [1][2] - The event serves as a platform for technology exchange and validation, showcasing advancements in humanoid robotics, which integrates artificial intelligence and deep learning for autonomous decision-making and adaptive capabilities [1][2] - The competition highlights the current early-stage development of humanoid robot technology, allowing companies and research institutions to demonstrate their progress and identify areas for improvement [1][2] Group 2: Industry Impact - The robot sports competition acts as a crucial indicator for the global humanoid robot industry, attracting 192 university teams and 88 corporate teams, fostering a high-density gathering of technology, talent, and capital [2] - The event assesses not only individual robot performance but also the maturity of the entire ecosystem, including algorithm quality, hardware supply chain stability, and operational response speed, driving improvements across the industry [2] - Investors gain direct insights into the practical applications and market potential of the technology, while government entities can better understand industry trends to formulate targeted policies that support development [2] Group 3: Social Influence - The competition builds a bridge between the public and cutting-edge technology, providing a unique platform for 16 countries to compete, which enhances public understanding of humanoid robots [3] - Spectators can directly observe the advancements and potential of robot technology, which could lead to exponential growth in product orders once market expectations are met [3] - The competitive atmosphere and recognition system encourage young talent to engage in robotics research, accelerating talent cultivation and accumulation in the industry [3]

划重点： 2025年7月1日，Meta正式宣布成立 Meta超级智能实验室（Meta Superintelligence Labs），以加速通用人工智能（AGI）的研发进程。在 这一重组过程中，马克·扎克伯格展现出对AI顶尖人才的强烈渴求，他主导的一系列高调挖角行动迅速震动了整个科技行业。然而，在此 过程中，长期担任Meta基础AI研究实验室（FAIR）首席科学家的杨立昆（Yann LeCun）却逐渐被"边缘化"，淡出了公众视野。 自2023年夏季以来，Llama模型家族已被下载约8亿次，这个数字令人震惊。 让AI的行为与人类价值观保持一致，更像是一个工程与设计问题，就像当年通过工程手段让喷气式飞机安全飞行一 样。 杨立昆提出"目标驱动架构"，核心思路是为系统设定清晰的目标和必要的安全边界，使其在既定范围内执行任务。 AI可能像15世纪的印刷术一样，引发一场新的文艺复兴，放大人类智慧。 年轻人不要被负面的、耸人听闻的故事吓倒；要认识到自己的力量，主动塑造自己想要的未来。 沉寂数月后，这位被誉为"人工智能教父"之一的科学家，在法国巴黎接受一位人工智能专家（也是硅谷一家初创公司的联合创始人、首 席技术官）——芭芭 ...