[AI Leadership] ③ Samsung’s AI Technique Centered on Buyer Experiences – Samsung World Newsroom

AI itself isn’t the Holy Grail at Samsung Electronics. Reasonably, AI is a software — a comparatively new and highly effective one — to make significant enhancements in customers’ on a regular basis lives. Constructing on a protracted custom of technological management and open innovation, Samsung leverages its numerous capabilities — that vary from operational applied sciences comparable to on-device and cloud AI to methods together with proprietary AI mannequin growth and partnerships — to ship sensible options and elevate the shopper expertise.

On this ultimate installment of Samsung’s AI Management collection, Samsung Newsroom explores how Samsung’s technological experience and analysis initiatives intention to refine AI-driven person interactions throughout its expansive portfolio.

Prioritizing Buyer Experiences

Samsung makes use of on-device and cloud-based AI options in its choices. This technique is additional strengthened by the pursuit of hybrid AI, which integrates the strengths of each applied sciences to offer context-aware, tailor-made options for numerous person wants. The choice on which sort or mixture of applied sciences to make use of is anchored by this final objective — “How can we finest serve the buyer?”

On-device AI delivers safe and environment friendly AI experiences in gadgets comparable to smartphones and TVs by using solely the info and sources on the machine itself. As a result of on-device AI doesn’t talk with the cloud, responses are nearly instantaneous. Since information doesn’t depart the machine, the know-how is usually thought-about safer for delicate procedures comparable to those who contain private data.

Equally, edge AI operates independently of cloud infrastructure and depends solely on particular native sources to carry out AI features. In interconnected house environments the place home equipment and IoT gadgets work in tandem, information processing in edge AI might be dealt with by both gadgets with increased efficiency specs or gadgets close by — with out relying on the cloud. Thus, edge AI presents pace and safety advantages akin to on-device AI.

Whereas on-device and edge AI depend on applied sciences that use smaller fashions and fewer computations in constrained {hardware} environments, experiences that require entry to intensive on-line data or high-performance computing are supported by cloud-based AI. The know-how makes use of huge datasets and large-scale AI fashions hosted on servers to ship a spread of superior capabilities. To attain excessive efficiency whereas sustaining effectivity and safety, Samsung conducts intensive analysis in areas together with minimizing information latency, optimizing mannequin effectivity and implementing information anonymization and encryption.

With experience in all these totally different types of AI — on-device, edge and cloud-based — Samsung takes a step additional and combines totally different codecs as essential to finest serve the wants of every use case. In hybrid AI, delicate and quick-response information might be processed utilizing on-device AI fashions whereas the cloud can handle processing that requires exterior sources like up-to-date data or high-performance fashions. Galaxy AI, for example, makes use of each on-device AI and cloud AI — working both independently or concurrently as wanted — to ship optimized options for customers.

▲ Daehyun Kim, Head of World AI Heart, Samsung Analysis explains Samsung’s AI Technique.

Generative AI Built-in Into On a regular basis Life

Samsung continues to drive innovation within the realm of generative AI, additional enhancing the flexibility of its services and products. As an illustration, Galaxy AI presents a set of options¹ together with Chat Help, which helps translation and interpretation in choose messaging apps; Notice Help, which helps summarize notes and generates be aware covers; and Photograph Help, which edits photographs with generative AI.

Generative AI can also be supported on choose Samsung TVs and residential home equipment by means of the Generative Wallpaper function.² Customers can take pleasure in 4K wallpapers that match their tastes on the 2024 Neo QLED TV or remodel their kitchens with AI-generated cowl screens on the Household Hub™ fridge.

▲ Generative Wallpaper on Samsung’s AI TV

To supply the perfect generative AI expertise on its gadgets, Samsung not solely collaborates with companions but in addition develops its personal AI fashions. From basic applied sciences starting with preliminary analysis — together with basis mannequin growth — to information assortment and coaching in addition to light-weight AI mannequin growth, Samsung leverages its huge experience throughout the AI R&D board to optimize options and meet every product’s particular wants.

For instance, Samsung developed proprietary basis fashions particularly for picture technology, enhancing and transformation to help a variety of visible creation experiences. This was a problem as a result of not like textual content information — which consists of discrete values — picture information consists of steady values, providing almost limitless potentialities for technology. To deal with this complexity, the corporate utilized specialised coaching strategies to the picture fashions whereas decreasing their dimension for environment friendly on-device operation. Builders refined deep studying algorithms and information augmentation strategies to permit for fast computation and seamless picture creation with minimal processing necessities.

These basis fashions had been then fine-tuned for particular functions. As an illustration, the Photograph Ambient wallpaper in Samsung’s newest smartphones integrates the picture mannequin with real-time information to create dynamic wallpapers that visually mirror present circumstances comparable to snowfall or a starlit evening relying on the time and climate.

Samsung Gauss2: Samsung’s Proprietary Generative AI

The appliance of generative AI is quickly increasing past private every day use into enterprise environments. Samsung has developed a proprietary generative AI mannequin to strengthen its competitiveness in AI know-how, securely handle delicate inside data and optimize efficiency and scalability for particular functions. Proprietary generative AI permits the corporate to ascertain fashions tailor-made to a variety of services and products whereas enhancing inside productiveness in a safer atmosphere.

Samsung lately unveiled its upgraded proprietary generative AI mannequin, Samsung Gauss2 with higher efficiency and effectivity in comparison with its predecessor, Samsung Gauss1, launched final yr. Samsung Gauss2 is a multimodal³ mannequin able to concurrently processing and understanding varied information varieties.

Samsung Gauss2 is offered in three fashions tailor-made to particular use circumstances — Compact, Balanced and Supreme. The Compact mannequin is a light-weight variant optimized for on-device AI. The Balanced mannequin, a medium-sized cloud-based variant, is designed for duties comparable to textual content and code technology. The Supreme mannequin, a big cloud-based variant, is constructed to help high-performance providers with a selected give attention to enhancing office productiveness.

▲ Samsung unveiled its upgraded proprietary generative AI mannequin, Samsung Gauss2.

To effectively implement giant fashions just like the Supreme variant, Samsung has included the Combination of Consultants (MoE) method to selectively activate solely the “professional fashions” best suited for particular duties inside a large-scale neural community.

MoE considerably reduces the variety of computations throughout coaching and inference by activating solely the required parts, quite than operating all professional fashions concurrently in response to every question. This selective activation not solely enhances effectivity but in addition facilitates the growth of the mannequin to deal with more and more complicated issues by including extra professional fashions. Samsung has meticulously skilled these professional fashions to function precisely and with out interference.

▲ MoE optimizes computation to implement giant AI fashions extra effectively.

Samsung Gauss2 is already being utilized throughout a number of areas inside Samsung. As an illustration, the Machine eXperience (DX) Division employs Samsung Gauss2’s coding assistant service to help analysis and growth actions. This service leverages a specialised mannequin — skilled on Samsung’s proprietary information, code and processes — to generate code wanted for product and repair growth. Totally built-in into the corporate’s inside software program growth atmosphere, the coding assistant helps pure language interactions and varied programming languages to streamline developer workflows.

The Samsung Gauss Portal, a conversational AI service, is broadly used for office duties comparable to doc summarization and translation, electronic mail drafting and extra. Future updates intention to reinforce the portal’s pure language question-and-answer capabilities and broaden its multimodal options to incorporate decoding tables and charts and producing photographs.

Strengthening AI Capabilities Via Shopper Engagement

Leveraging its intensive product strains, Samsung is growing applied sciences that help on-device, cloud and generative AI to ship a definite aggressive edge in AI capabilities.

By connecting with customers by means of its merchandise, the corporate positive factors precious insights into person behaviors and must allow the event of sensible AI options. This creates a virtuous cycle — whereby person information informs AI developments, which in flip ship worth again to customers.

Samsung {couples} this understanding of merchandise with in-house developments and strategic partnerships to implement optimum AI applied sciences. As an illustration, neural processing items (NPUs) — specialised accelerators for AI computation — should be tailor-made to satisfy the precise necessities of particular person merchandise, quite than utilized uniformly throughout all gadgets. Samsung not solely designs NPUs in-house for its personal merchandise but in addition applies software program optimizations to companion chipsets, making certain help for AI features and peak efficiency in Samsung gadgets.

As well as, the corporate attracts on the experience and views of product-specific specialists to deal with safety points and facilitate the fast utility of options to merchandise whereas driving steady technological enchancment. The blockchain-based, multi-device safety resolution Knox Matrix is a notable instance of how this collaborative method can create an atmosphere through which related gadgets verify one another for threats to reinforce safety.

▲ Samsung is main in AI due to shopper insights gained from varied merchandise.

Via the implementation of personalized AI applied sciences, Samsung continues to unlock new potentialities for customers. Trying forward, the corporate plans to ship extra differentiated AI experiences that empower customers by means of significant and enriching interactions.

¹ Availability of Galaxy AI options could differ by area, nation, OS or One UI model, mannequin and provider.
² Generative Wallpaper function could differ by area and mannequin.
³ Able to concurrently understanding and processing varied forms of information (modalities) together with textual content, photographs, audio and video.