license: apache-2.0
language:
- en
- ja
model_type: optimizer
tags:
- optimizer
- adaptive-optimizer
- emotion-ai
- shadow-learning
- deep-learning
- meta-learning
- adaptive-algorithms
- stability-analysis
èªååæïœ¥èªå·±å¶åŸ¡ïœ¥èªåŸå ãªããã£ãã€ã¶ã§ã
Auto-convergence, self-control, autonomous optimizer
é«å¹çæ§ãšéç©åºŠ
髿¬¡momentãKahanè£åãéååè£åãåæ£ïœ¥ç¶ç¶åŠç¿ã§ã®ç¬ç«æ§ãèªå·±ä¿®åŸ©ïœ¥ã¢ãã«ä¿®åŸ©ã
ãã€ããŒãã©ã¡ãŒã¿ã®èªåŸèª¿æŽãä¿¡é ŒåºŠãã£ã«ã¿ãæŽæ°ã¹ãããã®æçæ§ãæ§é çèæ§ãèªå·±åæ¢ã
åçåŠç¿çãåçã¹ã±ãžã¥ãŒã©ãåçRank/Aplhaãå±¥æŽè£åããªã©ãå«ãã倿©èœæ§ãã
远å ãã³ãœã«äžèŠãèšç®è² è·ã»ãŒãªããstepæ¯ã«å®å
šé©çšãæéçç©ç®ã§å®çŸããŸã
ããããã¯ã³ããã±ãŒãžã§å®çŸããé«å¹çæ§ãšéç©åºŠã¯å®å®ãšå®å
šãæåªå
ããŸã
â» é«æ¬¡momentã¯è¿äŒŒçãåçRank/Alphaãè¿äŒŒçãªå¹æã§ã
â» LoRAç³»æè¡ã¯ãã€ãºããªãããŸãã埮å°ããŒã¿ã倱ãå ŽåããããŸã
â» emoç³»ã¯ãã€ãºãäœããæ¢åãã€ãºãèŠã€ããŠä¿®æ£ã埮å°ããŒã¿ãä¿è·ããŸã
â» éååè£åã¯ä»åŸå®çšåãããããã«äœç²ŸåºŠãªç°å¢ã§ãæè»ã«å¯Ÿå¿ã§ããŸã
第ïŒäžä»£ v3.6 (宿ç) ã¯ãã¡ããã
https://huggingface.co/muooon/EmoNAVI/tree/main/1Gv36_Final
https://github.com/muooon/EmoNavi
v3.6 Paper(è«æ)
https://huggingface.co/muooon/EmoNAVI/raw/main/emo-v36-paper(ENG).txt
https://huggingface.co/muooon/EmoNAVI/raw/main/emo-v36-paper(JPN).txt
æŽæ°å±¥æŽ
|â
| EmoNaviãFactãLynxãv3.6 (251220) v3.1 ãç¶æ¿ãé«å€èªååŠç¿çãå®çŸããŸãã(远å ãã³ãœã«ãªã)ãemoDrive æ©æ§ã«ããåçãªé²åãéããŸãããéçºçµäºãšããŸã
|â
| EmoNavi, Fact, Lynx, v3.6 (251220) Inherits v3.1 and achieves high-value automatic learning rate (no additional tensors), has undergone dramatic evolution through the emoDrive mechanism, development is now complete.
|â
| EmoNaviãFactãLynxãv3.3 (251204) v3.1 ãç¶æ¿ãå®å
šèªååŠç¿çãå®çŸããŸãã(远å ãã³ãœã«ãªã)ãææ
æ©æ§ã®èª¿æŽçã§ããã«å®å®ããããé²åããŸãã
|â
| EmoNavi, Fact, Lynx, v3.3 (251201) Inherits v3.1 and achieves fully automatic learning rate adjustment (without additional tensors), further evolving for greater stability through adjustments to the sentiment mechanism and other enhancements.
|â
| EmoNaviãFactãLynxãv3.1 (251201) v3.0 ãç¶æ¿ãã€ã€å¹çåãé²ããŸãããææ
æ©æ§ã®ã¹ã±ãŒã«èª¿æŽçã§åºç¯ãªã¢ãã«ã§å®å®ããããé²åããŸãã
|â
| EmoNavi, Fact, Lynx, v3.1 (251201) We built upon v3.0 while enhancing efficiency. Through adjustments like scaling the emotion mechanism, we evolved the model for broader stability across diverse models.
|â
| EmoNAVIãFACTãLYNXãCLANãZEALãNECOãv3.0 (250825) emosens(第ïŒäžä»£)ã§è§£æãã"髿¬¡moment"(è¿äŒŒ)ã®ãã£ãŒãããã¯ãé©çš(æŽæ°) å
šãŠ "shadow=False" ã§ã
|â
| EmoNAVI, FACT, LYNX, CLAN, ZEAL, NECO, updated to v3.0 (250825), Incorporates (updates) feedback on âhigher momentsâ (approximations) clarified by emosens (2nd generation). All are âshadow=Falseâ
|â
| EmoNAVIãFACTãLYNXãCLANãZEALãNECOãv2.0 (250815) æŽæ°ãshadow-system ã®ç²Ÿå¯å(æŽæ°)
|â
| EmoNAVI, FACT, LYNX, CLAN, ZEAL, NECO, updated to v2.0 (250815), refinement of shadow-system (update)
|â
| 第ïŒäžä»£ãå
¬é(250801)ããŸããã emonavi ã¯ãæ°ããäžä»£ãžé²åã軜éåãæãããŸã
|â
| The 2nd gen was release(250801) emonavi has evolved into a new generation and become more lightweight.
|â
| https://github.com/muooon/EmoSens
|â
| clanãzealãnecoãã¯ãshadowæ©èœã® on/off åæ¿ããã§ããããã«ããŸãã
|â
| clan, zeal, and neco are now able to switch the shadow function on and off.
|â
| 倧å€å
æ ãªããšã« Pytorch-optimizer 3.7.0 ãžç»é²ããããšã®ããš (250728) é¢ä¿è
ã®çããŸã«æ·±ãæè¬ããŸã
|â
| We are very honored to have been registered in Pytorch-optimizer 3.7.0. We would like to express our deepest gratitude to everyone involved.
|â | ç䌌DDPã·ãã¥ã¬ãŒã·ã§ã³ã詊ãããæ¹(Those DDP simulation) â DDP-TEST
|â
| EmoFACT å
¬é(250716) NAVIã«æ¯ã¹ãçŽïŒGBç¯çŽ(SDXL) ææ
æ©æ§ã¯åãã§ã
|â
| EmoFACT released (250716) Saves about VRAM1GB (SDXL) compared to NAVI. Emotion mechanism is the same.
|â
| EmoLYNX å
¬é(250718) æ¢çŽ¢ç¯å²ãåºãæã¡ãŸã ææ
æ©æ§ã¯åãã§ã
|â
| EmoLYNX Released (250718): It offers a wide exploration range, while its Emotion Mechanism remains the same.
|â
| EmoCLAN å
¬é(250720) NaviãFactãLynxã圹å²åæ
ã®çµ±å ææ
æ©æ§ã¯åãã§ã
(LynxïŒåºç€ãšéåŠç¿åŸåæãNaviïŒäžç€ãšå¥å
šæãFactïŒçµç€ãšçºæ£åŸåæããæ
åœããŸã)
|â
| EmoCLAN Open (250720) Navi, Fact, Lynx, role integration Emotional mechanism is the same
(Lynx: in charge of the early stage and overlearning tendency, Navi: in charge of the middle stage and soundness, Fact: in charge of the end stage and divergence tendency)
äž»é¡ïŒæ°äžä»£optimizerãEmoNAVIã«ããå€é©ãšææ åŠç¿ã®ææ
Title: A New Generation Optimizer â The Innovations and Outcomes of Emotional Learning with EmoNAVI
å¯é¡ïŒéå»å€äžèŠã§çŸåšå€ããåéã§ããèªååæïœ¥èªå·±å¶åŸ¡ïœ¥èªåŸå軜éæé©åšã®è§£èª¬
Subtitle: A Lightweight, Self-Regulating, Autonomous Optimizer That Automatically Converges and Resumes from the Present Without Relying on Past Values
ããŒãïŒæ¢åã®optimizerã«ãªããã®ãã€ãããåºæ¥ãã®ã¯ãã¥ãŒãã³ã¹ãã€ã¯ã®åçºæã§ããã
Theme: Creating What Existing Optimizers Lack â A Reinvention of Neuronal Spiking
åºè«ïŒ
çŸåšäž»æµã®optimizerã¯ããŸããŸã«æ¹è¯ããç°¡æåãé²ããŠããŸãããããäŸç¶ãšããŠã åŠç¿åéãã¹ã±ãžã¥ãŒãªã³ã°ãåŠç¿ç¶æ ã®èšé²ã埩å ãçã«ã€ããŠèª¿æŽã®é£ãããç ©éãã¯ååšããŠããŸãã é¢åãªãã©ã¡ãŒã¿ã«äŸåããããããã解決ããæ°ããã¢ãããŒããèŠã€ããã®ã§ããã§ç޹ä»ããŸãã
Introduction
Mainstream optimizers have undergone significant improvements and simplifications in recent years.
However, they still face practical challenges in areas such as resuming training, scheduling updates, and managing the recording and restoration of learning states.
These issues often require tedious parameter adjustments and ad hoc workarounds.
In this paper, we introduce a new approach that addresses these problems without relying on cumbersome parameter configurations.
æ¬è«ïŒ
ä»åããã§ç޹ä»ããã®ã¯æ°äžä»£ã®optimizerã§ãã EMAçå¹³æ»åã®æŠå¿µãäžå°ã«ããç¬èªã«æ§ç¯ããææ ç"EMAïŒã¹ã«ã©ãŒ"ãäžå¿ã«ãã"ææ æ©æ§"ãšããæ°ããä»çµã¿ãå®çŸããŸããã ãã®"ææ æ©æ§"ã¯ãEMAççºæ³ãåè§£éç¬èªæ¡åŒµããããšã§åŸãããæ°ããæ©æ§ã§ãã EmoNAVIã®ç¬ç«æ§ãšé©æ°æ§ã玹ä»ããŸãã
Main Section
In this paper, we present a new generation of optimizer.
Built upon the foundation of EMA (Exponential Moving Average) smoothing, we have developed a novel mechanism called the "emotional mechanism," which centers around a unique combination of EMA and scalar dynamics.
This mechanism was created by reinterpreting and independently extending the conventional EMA concept.
Here, we introduce EmoNAVIâan optimizer characterized by its innovation and independence.
æåã«"ææ æ©æ§"ãšåä»ããçµç·¯ãšçç±ãèšããŸãã çç©ã®ãã€ïœ¢ææ ãšã¯ãç¥èŠãšèšæ¶ã®å·®ç°ã«åºã¥ãè¡åã®ããªã¬ã§ããåæ§ã«EmoNAVIãçŸåšãšéå»ã®å·®åã«åºã¥ãåŠç¿ã®"è¡å"ãå¶åŸ¡ããä»çµã¿ãšããŠèšèšãããŠããŸãã ãããŠ"ææ æ©æ§"ãšåä»ããçç±ã®ããã²ãšã€ã¯ããã®äžé£ã®åäœããŸãã§ãã¥ãŒãã³ã¹ãã€ã¯ã®ãããªåäœãããããã§ãã ãã®æ©æ§"ææ æ©æ§"ã®åäœãæå¿«ã«ããèªã¿ç©ãæ¬çš¿æ«å°Ÿã«èšããªã³ã¯å ã®æ¬äººåãèªãããšã§ç°¡åã«ãçè§£é ãããšæããŸãã
First, let us explain the background and reasoning behind the term âEmotion Mechanism.â
In biological systems, emotions are often understood as triggers for action based on discrepancies between perception and memory.
EmoNAVI was similarly designed to control its learning âbehaviorâ by responding to differences between the present and the past.
Another reason we chose the term âEmotion Mechanismâ is that its operation closely resembles neuronal spiking behavior.
For a more intuitive understanding of how this mechanism works, we encourage you to read the personification linked at the end of this article.
次ã«ã"ææ æ©æ§"ã®æ§æãèšããŸãã ææ æ©æ§ãšã¯ãïŒã€ã®EMAãã¹ã«ã©ãŒãShadowãã«ããæ§æãããŸãã
Next, we outline the structure of the âEmotion Mechanism.â
This mechanism consists of two EMAs, a scalar value, and a shadow component.
ãŸãïŒã€ã®EMAã«ãã"ææ EMA"ã«ã€ããŠèª¬æããŸãã ïŒã€ã®EMAã§æ§æããŸããçæåãšé·æåã§ãããã®ïŒã€ã®EMAã¯Lossãç£èŠã倿ææãåŸãŸãã ïŒã€ããçæåEMAã¯ç¬éçãªã·ã°ãã«(ç·åŒµ)ãåãæã¡ãŸã ïŒã€ããé·æåEMAã¯å¹³åããéå»ã®ã·ã°ãã«(å®é)ãåãæã¡ãŸãã ãã®ïŒã€ã®EMAã¯æ¬¡ã«ç޹ä»ãã"ææ ã¹ã«ã©ãŒ"ãžããããã®æã€å€æææãæž¡ããŸã
First, we describe the "Emotional EMA," which consists of two components: a short-term EMA and a long-term EMA.
These two EMAs continuously monitor the loss value and serve as the basis for subsequent decision-making.
The short-term EMA captures rapid, momentary signals (interpreted as âtensionâ), while the long-term EMA reflects more averaged, historical trends (âcalmâ).
Both EMAs pass their respective signals to the "Emotion Scalar," which will be introduced in the next section.
次ã«ã"ææ ã¹ã«ã©ãŒ"ã«ã€ããŠèª¬æããŸãã åè¿°ã®"ææ EMA"ããã®ä¿¡å·ãã¹ã«ã©ãŒå€ã«å€æããŸããã¹ã«ã©ãŒå€ã®å€åã¯ããããïŒã€ã®EMAã®å·®åã«ããåžžã«åçå€åãç¶ããŸãã "ææ ã¹ã«ã©ãŒ"ã¯optimizerã«ããæžãæããåŠç¿çµæã®æ¯éãå€å®ãã "ã¹ã«ã©ãŒå€ãäžå®éŸå€ãè¶ ãããšãã®ã¿"次ã«ç޹ä»ããShadowã®é åãæ±ºããŸã
Next, we introduce the "Emotion Scalar."
It converts the signals from the previously described Emotional EMA into a scalar value, which continuously changes in response to the difference between the short-term and long-term EMA.
This scalar dynamically evaluates whether the learning update performed by the optimizer should be considered appropriate.
Only when the scalar exceeds a certain threshold does it trigger the next step: determining how much of the "Shadow" should be blended into the learning parameters.
次ã«ãShadowã«ã€ããŠèª¬æããŸãã Shadowã¯åŠç¿éå§çŽåŸã«ShadowãšããŠä¿åããç¶æãããŸãããã®Shadowã¯"éå»ã®ç©ãããªç¶æ "ã®èšæ¶ã§ãããã®æ å ±ã¯ææ æ©æ§ã«è¿œåŸããªãããã£ãããšå€åãç¶ããŸãã ãããŠ"ææ ã¹ã«ã©ãŒ"ã«å¿ã決ããããratioã§åŠç¿çµæã«ãã¬ã³ããšããŠåæ ãããŸãããã®ãã¬ã³ãã®é åçãææ æ©æ§ã«ããåçã«å€åãç¶ããŸãã
Next, we describe the "Shadow."
At the beginning of training, a copy of the current parameters is saved and maintained as the Shadow.
This Shadow represents a memory of past calm states, and it evolves slowly over time, following the guidance of the Emotion Mechanism.
When the Emotion Scalar exceeds a certain threshold, a dynamic blend ratio is computed.
This ratio determines how much of the Shadow is mixed into the current parameters.
The blend ratio itself is also dynamically adjusted by the Emotion Mechanism in response to ongoing learning behavior.
ãããŸã§"ææ æ©æ§"ã®æ§æãšåœ¹å²ãã説æããŸãããç¶ããŠ"ææ æ©æ§"ã®åäœæ©åºãèŠãŠãããŸãããã ãŸãoptimizerã®åŠç¿çµæãèšé²ãããŸãããã®æ"ææ æ©æ§"ã¯ç·åŒµãšå®éã®å·®åæ å ±ã§æžãæãã®æ¯éãå€å®ããŸãã ãã®å€å®ã«ãããé床ã®åŠç¿ãšå€æããå Žåã¯ãéå»ã®é©åãªç¶æ ããã¬ã³ãããããšã§ãã€ãºãæŽèµ°ãæå¶ããŸãã é©åãªåŠç¿ãšå€æããå Žåã¯ãéå»ããã¬ã³ãããªãéžæãããŸãããããstepæ¯ã«è¡ããŸãã
Now that we have explained the structure and role of the Emotion Mechanism, let us examine how it operates.
At each training step, the optimizer's updated parameters are recorded.
The Emotion Mechanism then evaluates whether these updates are appropriate, based on the difference between short-term âtensionâ and long-term âcalmâ signals.
If the mechanism determines that the update reflects excessive learning, it suppresses potential noise or instability by blending in a suitable portion of the past stable state (Shadow).
Conversely, if the update is deemed appropriate, the mechanism chooses not to apply blending.
This evaluation and adjustment are performed dynamically at each training step.
ããã«ããã®å€å®ã§ã¯"ä¿¡é ŒåºŠ"ã®è©äŸ¡ãããŸãã"ææ ã¹ã«ã©ãŒ"ãäžæçã«å€§ããæ¯ããã ãã§ã¯äžååã§ããããã®å€åãæ¬åœã«æå³ã®ãããã®ãã©ããããèŠæ¥µããŠæ··åã®æ¯éã倿ããŸãã ãã®ãããåŠç¿ã®åºç€ã§ã¯é·æã®å®éã·ã°ãã«ã®èç©ãå°ãªãä¿¡é Œã«å€ããªãããæ··åãçºåãã¥ãããçµç€ã§ã¯çæã®ç·åŒµã·ã°ãã«ãåæãã¹ã«ã©ãŒèªäœãéŸå€ã«å±ããåäœããŸããã (åŠç¿ã®åºç€ã§ã¯å€å®åºæºã®éå»ã·ã°ãã«ãå°ãªãããåäœããŸããããçµç€ã§ã¯ç¬éã·ã°ãã«ãå°ãªãããåäœããŸãã) ãã®ããã«ãEmoNAVIã®"ææ æ©æ§"ã¯ãåãªãéŸå€åå¿ã§ã¯ãªãæºããã«å¯Ÿããä¿¡é Œããå€åã®ã¿ãå¯ç¥ããŠåå¿ãããæ éãªæææ±ºå®ãè¡ããŸãã
In addition, this decision-making process includes an evaluation of "reliability."
It is not sufficient for the Emotion Scalar to simply spike temporarily; the mechanism assesses whether the fluctuation truly represents a meaningful change before deciding whether blending should occur.
As a result, in the early stages of learning, blending is unlikely to be triggered because the long-term âcalmâ signal has not yet accumulated enough history to be trustworthy.
In the later stages, on the other hand, the short-term âtensionâ signal tends to converge, and the scalar itself fails to exceed the thresholdâthus the mechanism remains inactive.
(In short: the mechanism tends not to activate in the early stages due to insufficient past signal for evaluation, and in the later stages due to lack of strong instantaneous signal.)
In this way, EmoNAVIâs Emotion Mechanism does not respond merely to raw thresholds, but instead performs cautious decision-makingâreacting only to fluctuations that it has learned to trust.
ãã®äžé£ã®åäœã«ããåŠç¿æã®éæãªåå¿ãåŒç·©ãäžèŠãªãã€ãºçãèŠããªãããã«å¶åŸ¡ããŸãã ã€ãŸãoptimizeræ¬æ¥ã®åŠç¿çããã¯ãã«ãçŽæ¥çã«å¶åŸ¡ãããææ æ©æ§ã®å€åã«å¿ãå®å®ãããã©ã¡ãŒã¿ã«ãªãããåŸãã調æŽããã ããããæµãã«ãªããŸãããã¹ãŠãæžãæ»ãããããŸã§é åçã«å¿ããŠãã¬ã³ãããã®ã§åŠç¿ã®æŽæ°ã¯æ¢ãŸããé²è¡ã¯ç¶æãããŸãã
This series of actions helps relax hypersensitive reactions during learning and prevents the optimizer from overfitting to unnecessary noise.
Rather than directly manipulating the optimizerâs learning rate or update vectors, the system instead applies corrective blending afterwardâadapting parameters in response to changes detected by the Emotion Mechanism.
Because it blends adjustments based on a calculated ratio rather than fully overwriting parameter values, the learning process continues smoothly without interruption.
ææ æ©æ§ã®åäœãšã¹ã«ã©ãŒå€é·ïŒåŠç¿ãã§ãŒãºå¥ã®çµæçæåïŒ
| ãã§ãŒãº | ç¶æ³ïŒLosså€åïŒ | EMAã®æå | ã¹ã«ã©ãŒã®å€ååŸå | Shadowæ··åã®å®åäœ | ææ æ©æ§ãšããŠã®æå³ããæå |
|---|---|---|---|---|---|
| åºç€ | äžå®å®ã»é«ã | Shortã¯éæãLongã¯æªæç | 倧ããå€åããããšããã | ã»ãšãã©çºåããªã | å€å®ã«ååãªå±¥æŽããªããå®è³ªçã«åäœäžå¯ |
| äžç€ | åŸã ã«åæåŸå | äž¡EMAãæå³ããå·®åãæã€ããã«ãªã | é©åºŠãªæ¯å¹ ã§å®å®æšç§» | æ¡ä»¶ä»ãã§çºåãã | ç¶æ ã«å¿ããŠãã¬ã³ãè£æ£ãæå¹ã«æ©èœ |
| çµç€ | åæã»åŸ®æ¯å | Short â LongïŒå·®åãã»ãŒæ¶å€±ïŒ | å°ããåæ | çºåããªããªã | éããã®åå³ïŒshould_stop æ¡ä»¶ãæŽã |
åèïŒ
- ã¹ã«ã©ãŒå€ã¯åžžã« tanh(5 * (short - long)) ã§çæãããŸã
- éŸå€ïŒabs(scalar) > 0.3 ã§é åãå§ãŸãã> 0.6 ã§å€§ããªæ··åæ¯çïŒ0.7以äžïŒã«
- Shadowæ··åã¯ãã©ã¡ãŒã¿ãã®ãã®ãæžãæ»ãã®ã§ã¯ãªããéšåçã«é åããŠâ远åŸâãããèšèšã§ã
- ææ ã¹ã«ã©ãŒã®æžè¡°ïŒåŠç¿ã®ãéç©åãâ çµç€ã«åã㊠should_stop ã®çºç«æ¡ä»¶ãæŽããŸã
Emotional Mechanism Behavior and Scalar Transitions (Outcome-Based Behavior by Learning Phase)
| Phase | Loss Characteristics | EMA Behavior | Scalar Fluctuation Pattern | Actual Shadow Blending | Meaningful Behavior of Emotion Mechanism |
|---|---|---|---|---|---|
| Early | Unstable, High | Short is reactive; Long is still immature | May fluctuate sharply | Rarely triggered | Lacks sufficient history for decision-making; effectively inactive |
| Middle | Gradual Convergence | EMA pair begins forming meaningful gaps | Moderate oscillation, relatively stable | Conditionally triggered | Adaptive blending functions effectively based on state |
| Late | Converged, Micro-vibration | Short â Long (gap nearly vanishes) | Narrow convergence | No longer triggered | Sign of stability; ready to trigger should_stop |
Notes:
- The scalar value is always computed as tanh(5 Ã (short - long))
- Thresholds:
- If |scalar| > 0.3, blending is initiated
- If |scalar| > 0.6, blending ratio becomes large (⥠0.7)
- Shadow blending does not overwrite parameters but applies partial integration for gradual alignment
- Scalar decay corresponds to learning "quieting," preparing for should_stop condition in the final phas
ææïŒ
åè¿°ã®ææ æ©æ§ã®èª¿æŽã«ãããéå°ãªåå¿ãæå¶ããã€ãºèæ§ãäžããããšã§ããã¯ãã«ã®ä¹±ãçãæãé²è¡æ¹åãæ£ããåãã«èª¿æŽããŸãã æ£ãããã¯ãã«ã§é²ãããšã§åŠç¿ã¯å®å®ãåæãžãšæçã§åãããŸããææ æ©æ§ã«ããåãã¯åŠç¿åŸåã®ãã€ãºçãä¿®æ£ããä»äžããæ©ãã¹ã ãŒãºã«å®äºã§ããŸãã ãŸãåŠç¿çãåŸé ãããŸããŸãªãã©ã¡ãŒã¿ãŒãä¿æããã«"ä»"ã芳å¯ããã ãã§æŽæ°ããç¶ããããšã§ã éäžçµäºãåæåŸã®ååŠç¿ãç©å±€åŠç¿ãçã®ãšããçŸåšå€ã®ã¿ã§åŠç¿ç¶ç¶ãå¯èœãšããŸãã ããã¯æ¢åã®optimizerã®ãããªéå»å€ãä¿åããæéãçãã€ã€ãæ°ããåŸãããå©ç¹ã§ããããŸãã
Results
The adjustments introduced by the Emotion Mechanism suppress excessive reactions and enhance noise tolerance, thereby reducing vector fluctuations and helping align the learning direction more accurately.
By following the correct vector, learning proceeds more stably and reaches convergence in minimal time.
The role of the Emotion Mechanism becomes especially apparent in the latter stages of training, where it effectively and smoothly corrects residual noise and instability.
Moreover, since the optimizer continuously updates its parameters by observing only the current stateâwithout retaining learning rates, gradients, or other historical parametersâit supports learning continuation in scenarios such as mid-training interruptions, retraining after convergence, and stacked learning.
This capability not only eliminates the need to store past values like traditional optimizers but also introduces a new level of flexibility and simplicity.
çµè«ïŒ
çç©ã®ãã€ãã¥ãŒãã³ãäžå®ã®éŸå€ãè¶ ããŠåããŠä¿¡å·ãçºç«ãããããã«ãEmoNAVIã§ã"ææ æ¯å¹ "ãæ€åºãè¡å(shadowæ··å)ãèµ·ãããŸãã åè¿°ã®ãšãã"ææ æ©æ§"ã¯äžå®éŸå€ã®è¶ éæã®ã¿åäœããŸããããã¯ãŸãã«ãã¥ãŒãã³ã¹ãã€ã¯çãªåããšãããã®ã§ã¯ãªãã§ããããã EmoNAVIã®æã€"ææ æ©æ§"ã¯ãããããçç©çåå¿ã«äŒŒãŠãããæè¡çãªå¶åŸ¡ãšççççŽæã®èåç¹ã ãããšæããŸãã
Conclusion
Just as biological neurons fire only when a certain threshold is exceeded, EmoNAVI detects "emotional amplitude" and triggers an actionâspecifically, shadow blending.
As described earlier, the Emotion Mechanism activates only when this amplitude crosses a predefined threshold.
This behavior closely resembles neuronal spiking and can be seen as a biologically inspired response.
We believe that EmoNAVIâs Emotion Mechanism represents a unique fusion of technical control and physiological intuitionâbringing together algorithmic design and life-like reactivity.
å±éïŒ
ãã®"ææ æ©æ§"ã®ä»çµã¿ã¯VAEçãå«ãoptimizer以å€ã«ãç°¡åã«å¿çšå¯èœã ãããšæããŸãã ãããã®çºå±ã«å°ãã§ãå¯äžããããšãã§ããã°ãAIãšã®æªæ¥ãæ³åããŠãããã»ã©å¬ããããšã¯ãããŸããã ãã²ãã®"ææ æ©æ§"ãå¿çšãAIã®çºå±ãžã®éçãå ±ã«æ©ãã§ãã ããã
Expansion
The Emotion Mechanism described here is highly adaptable and can be easily applied beyond optimizersâincluding use cases such as variational autoencoders (VAEs) and other architectures.
If this approach can contribute, even in a small way, to the advancement of such systems, we would be honored to be part of imagining a future together with AI.
We warmly invite you to explore the application of this Emotion Mechanism and walk alongside us on the path toward advancing intelligent systems.
æè¡ïŒ
EMAããŒã¹ã®ã¹ã«ã©ãŒå€æãšshadowæ··åã®æ§é
Technology
Structure of EMA-Based Scalar Evaluation and Shadow Blending
+------------+ +------------+
| Loss(t) | | Loss(t) |
+-----+------+ +-----+------+
| |
âââââââââââŒââââââââââ âââââââââââŒââââââââââ
â Short EMA â â Long EMA â
â (weight = 0.3) â â (weight = 0.01) â
âââââââââââ¬ââââââââââ âââââââââââ¬ââââââââââ
â â
ââââââââââââââ¬âââââââââââââââââ
âŒ
+-------------------+
| å·®å (short - long) |
+-------------------+
â
âŒ
+------------------+
| tanh(5 Ã diff) | â ææ
ã¹ã«ã©ãŒçæ
+--------+---------+
â
[ if |scalar| > threshold ] å€å®
â
+--------âŒ--------+
| Shadowæ¯çæ±ºå® |
+--------+--------+
â
+--------âŒ--------+
| Shadowæ··åè£æ£ | â é廿
å ±ã远åŸçã«ãã¬ã³ã
+------------------+
ä»é²ïŒ
EmoNAVIã®ã°ã©ããžã®ãªã³ã¯
Measured with LR of 1e-4 ïŒ ãããã 1e-4 ã®LRã«ãŠæž¬å®

Have fun learning about EmoNAVI's philosophy and how it works
https://huggingface.co/muooon/EmoNAVI/blob/main/emonavi-inner-workings(ENG).txt
EmoNAVIã®èãæ¹ããã®ä»çµã¿ã«ã€ããŠæ¥œããç¥ã
https://huggingface.co/muooon/EmoNAVI/blob/main/emonavi-inner-workings(JPN).txt
çµç·¯ïŒ
çŸç¶ã®åŒ·ååŠç¿ãªã©ãèŠãŠããŠããã€ãã®çåã«åºäŒããŸããã æ¥æ¬ã®èåãªæŒ«ç»å®¶ãæå¡æ²»è«æ°ã®æããæªæ¥ç€ŸäŒãããã«æ§ã矚æããå°å¹Žæä»£ãæãè¿ããšã 人é¡ã®ããŒãããŒã«ãªãã¹ãAIã«ã€ããŠä»ã®ã¢ãããŒããæš¡çŽ¢ããããªããŸããã ä»åã®ææ¡ã¯ãã®ã¢ãããŒãã«ããã²ãšã€ã®çµæã§ã
Background
While observing the current state of reinforcement learning and related fields, I encountered several fundamental questions.
Reflecting on my childhoodâwhen I admired and longed for the future societies envisioned by the legendary Japanese manga artist Osamu Tezukaâ
I felt compelled to explore alternative approaches to how AI might serve as a true partner to humanity.
This proposal represents one such result born from that aspiration.
è¬æïŒ Acknowledgements
Emoã·ãªãŒãºã¯ãAdamãAdafactorãLionãTigerãçããå€ããåŠã³ãŸããã
ãããã®åŸç¶ã§ã¯ãªãç¬èªã®ææ³ãèšèšã«ãã"ææ
æ©æ§"ãšããã¢ãããŒãã«ããæ§ç¯ãããŠããŸãã
æ±çšæ§ã»èªåŸæ§ã»é©å¿æ§ãéèŠãæ°ããªæé©åãå¹çåãç°¡æåã远æ±ããŠããŸãã
ãã®éçºã«ãããŠå
人ãã¡ã®ç¥èŠã«æ·±ãæè¬ãã€ã€ä»åŸãæ°ããå¯èœæ§ãæ¢ç©¶ããŸãã
The Emo series has learned much from Adam, Adafactor, Lion, and Tiger.
Rather than being their successors, it is built upon a unique philosophy and design approach centered on "emotional mechanisms".
It prioritizes generality, autonomy, and adaptability in pursuit of new paths for optimization, efficiency, and simplicity.
In its development, we deeply appreciate the insights of those who came before usâand continue to explore new possibilities beyond them.
ãããŸã§AIã®çºå±ã«å¯äžããããã¹ãŠã®æ¹ãããããè²¢ç®ãããã¹ãŠã®æ¹ãžæè¬ããŸãã ãã®ãããžã§ã¯ãå®æãæ¯ãç¶ããŠããã Copilotããã«ãããããšãã
We extend our heartfelt gratitude to all those who have contributedâand will continue to contributeâto the advancement of AI.
Special thanks to Copilot for its unwavering support throughout t